<?xml version="1.0" encoding="utf-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" version="2.0"><channel><title>医工成果创新网</title><link>http://www.trizmed.cn/</link><description>医创助手</description><item><title>初识CUDA编程：理解线程索引与Warp分组</title><link>http://www.trizmed.cn/jscxd/19.html</link><description>&lt;ol class=&quot; list-paddingleft-2&quot; style=&quot;-w t, &amp;quot;Helvetica Neue&amp;quot;, &amp;quot;Hiragino Sans GB&amp;quot;, &amp;quot;Microsoft YaHei UI&amp;quot;, &amp;quot;Microsoft YaHei&amp;quot;, Arial, sans-serif; font-size: 17px; letter-spacing: 0.544px; text-align: justify; text-wrap: wrap; background-color: rgb(255, 255, 255); visibility: visible; box-sizing: border-box !important; overflow-wrap: break-word !important;&quot;&gt;&lt;li&gt;&lt;p style=&quot;-w g: border-box !important; overflow-wrap: break-word !important;&quot;&gt;线程索引与全局计算：CUDA中的基础定位机制&lt;/p&gt;&lt;/li&gt;&lt;/ol&gt;&lt;p&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent; margin: 0px; padding: 0px; outline: 0px; max-width: 100%; letter-spacing: 0.578px; visibility: visible; box-sizing: border-box !important; overflow-wrap: break-word !important;&quot;&gt;在CUDA编程中，每个线程都有一个唯一的全局索引，它由线程在块中的位置（&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent; margin: 0px; padding: 0px; outline: 0px; max-width: 100%; letter-spacing: 0.578px; visibility: visible; box-sizing: border-box !important; overflow-wrap: break-word !important;&quot;&gt;threadIdx&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent; margin: 0px; padding: 0px; outline: 0px; max-width: 100%; letter-spacing: 0.578px; visibility: visible; box-sizing: border-box !important; overflow-wrap: break-word !important;&quot;&gt;）以及块在网格中的位置（&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent; margin: 0px; padding: 0px; outline: 0px; max-width: 100%; letter-spacing: 0.578px; visibility: visible; box-sizing: border-box !important; overflow-wrap: break-word !important;&quot;&gt;blockIdx&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent; margin: 0px; padding: 0px; outline: 0px; max-width: 100%; letter-spacing: 0.578px; visibility: visible; box-sizing: border-box !important; overflow-wrap: break-word !important;&quot;&gt;）共同决定。&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent; margin: 0px; padding: 0px; outline: 0px; max-width: 100%; letter-spacing: 0.578px; visibility: visible; box-sizing: border-box !important; overflow-wrap: break-word !important;&quot;&gt;这个全局索引使得每个线程能够在并行计算中独立地访问和处理特定的数据位置，从而实现数据的分布式计算和处理。&lt;/span&gt;2. 块（Block）与网格（Grid）：分层并行计算的架构&lt;/p&gt;&lt;p style=&quot;-w t, &amp;quot;H g: border-box !important; overflow-wrap: break-word !important;&quot;&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent; margin: 0px; padding: 0px; outline: 0px; max-width: 100%; box-sizing: border-box !important; overflow-wrap: break-word !important; text-indent: 2em; font-size: var(--articleFontsize); letter-spacing: 0.034em; visibility: visible;&quot;&gt;CUDA将计算任务组织为多个块（blocks），每个块又包含若干线程。&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent; margin: 0px; padding: 0px; outline: 0px; max-width: 100%; box-sizing: border-box !important; overflow-wrap: break-word !important; text-indent: 2em; font-size: var(--articleFontsize); letter-spacing: 0.034em; visibility: visible;&quot;&gt;所有块组成一个网格（grid），并行执行在GPU上。&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent; margin: 0px; padding: 0px; outline: 0px; max-width: 100%; box-sizing: border-box !important; overflow-wrap: break-word !important; text-indent: 2em; font-size: var(--articleFontsize); letter-spacing: 0.034em; visibility: visible;&quot;&gt;这种分层结构让程序能够灵活地分配和管理大量并行线程，充分利用GPU的计算能力。&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent; margin: 0px; padding: 0px; outline: 0px; max-width: 100%; box-sizing: border-box !important; overflow-wrap: break-word !important; text-indent: 2em; font-size: var(--articleFontsize); letter-spacing: 0.034em; visibility: visible;&quot;&gt;块的索引（&lt;/span&gt;&lt;code style=&quot;-webkit-tap-highlight-color: transparent; margin: 0px; padding: 0px; outline: 0px; max-width: 100%; box-sizing: border-box !important; overflow-wrap: break-word !important; text-indent: 2em; font-size: var(--articleFontsize); letter-spacing: 0.034em; visibility: visible;&quot;&gt;blockIdx&lt;/code&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent; margin: 0px; padding: 0px; outline: 0px; max-width: 100%; box-sizing: border-box !important; overflow-wrap: break-word !important; text-indent: 2em; font-size: var(--articleFontsize); letter-spacing: 0.034em; visibility: visible;&quot;&gt;）和线程的索引（&lt;/span&gt;&lt;code style=&quot;-webkit-tap-highlight-color: transparent; margin: 0px; padding: 0px; outline: 0px; max-width: 100%; box-sizing: border-box !important; overflow-wrap: break-word !important; text-indent: 2em; font-size: var(--articleFontsize); letter-spacing: 0.034em; visibility: visible;&quot;&gt;threadIdx&lt;/code&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent; margin: 0px; padding: 0px; outline: 0px; max-width: 100%; box-sizing: border-box !important; overflow-wrap: break-word !important; text-indent: 2em; font-size: var(--articleFontsize); letter-spacing: 0.034em; visibility: visible;&quot;&gt;）共同用于在程序中定位和管理计算任务。&lt;/span&gt;&lt;/p&gt;&lt;p&gt;3. Warp：GPU中的基础执行单元&lt;/p&gt;&lt;p style=&quot;-w t, &amp;quot;H g: border-box !important; overflow-wrap: break-word !important;&quot;&gt;在CUDA架构中，warp是执行任务的最小单元，通常由32个连续的线程组成。所有的warp内线程同时执行相同的指令，通过优化warp内的指令调度，CUDA能够大幅提高并行计算的效率。理解warp的工作机制对于编写高效的CUDA程序至关重要，因为warp内的线程分支和内存访问模式将直接影响性能。&lt;/p&gt;&lt;p&gt;4. GPU内存管理：从CPU到GPU的数据传输&lt;/p&gt;&lt;p style=&quot;-w t, &amp;quot;Helvetica Neue&amp;quot;, &amp;quot;Hiragino Sans GB&amp;quot;, &amp;quot;Microsoft YaHei UI&amp;quot;, &amp;quot;Microsoft YaHei&amp;quot;, Arial, sans-serif; font-size: 17px; letter-spacing: 0.544px; text-align: justify; text-wrap: wrap; background-color: rgb(255, 255, 255); text-indent: 2em; box-sizing: border-box !important; overflow-wrap: break-word !important;&quot;&gt;在CUDA编程中，GPU内存管理是性能优化的关键因素。数据需要从CPU传输到GPU进行并行计算，然后再将结果传回。使用&lt;code style=&quot;-webkit-tap-highlight-color: transparent; margin: 0px; padding: 0px; outline: 0px; max-width: 100%; box-sizing: border-box !important; overflow-wrap: break-word !important;&quot;&gt;cudaMalloc&lt;/code&gt;、&lt;code style=&quot;-webkit-tap-highlight-color: transparent; margin: 0px; padding: 0px; outline: 0px; max-width: 100%; box-sizing: border-box !important; overflow-wrap: break-word !important;&quot;&gt;cudaMemcpy&lt;/code&gt;等函数，可以在GPU上动态分配内存和实现主机（CPU）与设备（GPU）之间的数据交换，确保程序在大规模数据并行处理中保持高效。&lt;/p&gt;&lt;p&gt;5. 同步与异步：CUDA中的并行执行模型&lt;/p&gt;&lt;p style=&quot;-w t, &amp;quot;Helvetica Neue&amp;quot;, &amp;quot;Hiragino Sans GB&amp;quot;, &amp;quot;Microsoft YaHei UI&amp;quot;, &amp;quot;Microsoft YaHei&amp;quot;, Arial, sans-serif; font-size: 17px; letter-spacing: 0.544px; text-align: justify; text-wrap: wrap; background-color: rgb(255, 255, 255); text-indent: 2em; box-sizing: border-box !important; overflow-wrap: break-word !important;&quot;&gt;CUDA程序中的内核函数（kernel）在设备上执行时，主机（CPU）和设备（GPU）之间是异步的，意味着CPU可以在不等待GPU完成任务的情况下继续执行其他操作。这种异步执行模式提高了程序的整体效率，但也需要程序员注意同步问题，如使用&lt;code style=&quot;-webkit-tap-highlight-color: transparent; margin: 0px; padding: 0px; outline: 0px; max-width: 100%; box-sizing: border-box !important; overflow-wrap: break-word !important;&quot;&gt;cudaMemcpy&lt;/code&gt;确保数据在正确的时间被传输和使用。&lt;span style=&quot;color: #AFAFAF; font-family: Consolas, &amp;quot;Liberation Mono&amp;quot;, Menlo, Courier, monospace; white-space: pre; background-color: #F8F8F8; font-size: 14px;&quot;&gt;#&lt;/span&gt;&lt;span style=&quot;color: #AFAFAF; font-family: Consolas, &amp;quot;Liberation Mono&amp;quot;, Menlo, Courier, monospace; white-space: pre; background-color: #F8F8F8; font-size: 14px; -webkit-tap-highlight-color: transparent; padding: 0px; outline: 0px; max-width: 1000%; box-sizing: border-box !important; overflow-wrap: break-word !important;&quot;&gt;include&lt;/span&gt;&lt;span style=&quot;color: #AFAFAF; font-family: Consolas, &amp;quot;Liberation Mono&amp;quot;, Menlo, Courier, monospace; white-space: pre; background-color: #F8F8F8; font-size: 14px;&quot;&gt; &lt;/span&gt;&lt;span style=&quot;color: #AFAFAF; font-family: Consolas, &amp;quot;Liberation Mono&amp;quot;, Menlo, Courier, monospace; white-space: pre; background-color: #F8F8F8; font-size: 14px; -webkit-tap-highlight-color: transparent; padding: 0px; outline: 0px; max-width: 1000%; box-sizing: border-box !important; overflow-wrap: break-word !important;&quot;&gt;&amp;lt;stdio.h&amp;gt;&lt;/span&gt;&lt;/p&gt;&lt;pre style=&quot;overflow-wrap: break-word !important;&quot; class=&quot;prism-highlight prism-language-cpp&quot;&gt;#include&amp;nbsp;&amp;lt;stdio.h&amp;gt;
#include&amp;nbsp;&amp;lt;stdlib.h&amp;gt;
#include&amp;nbsp;&amp;lt;stdio.h&amp;gt;
#include&amp;nbsp;&amp;lt;termios.h&amp;gt;
#include&amp;nbsp;&amp;lt;unistd.h&amp;gt;

int&amp;nbsp;getch(void)&amp;nbsp;{
&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;struct&amp;nbsp;termios&amp;nbsp;oldattr,&amp;nbsp;newattr;
&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;int&amp;nbsp;ch;
&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;tcgetattr(STDIN_FILENO,&amp;nbsp;&amp;amp;oldattr);&amp;nbsp;//&amp;nbsp;获取终端属性
&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;newattr&amp;nbsp;=&amp;nbsp;oldattr;
&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;newattr.c_lflag&amp;nbsp;&amp;amp;=&amp;nbsp;~(ICANON&amp;nbsp;|&amp;nbsp;ECHO);&amp;nbsp;//&amp;nbsp;禁用缓冲区输入和回显
&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;tcsetattr(STDIN_FILENO,&amp;nbsp;TCSANOW,&amp;nbsp;&amp;amp;newattr);
&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;ch&amp;nbsp;=&amp;nbsp;getchar();&amp;nbsp;//&amp;nbsp;获取字符输入
&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;tcsetattr(STDIN_FILENO,&amp;nbsp;TCSANOW,&amp;nbsp;&amp;amp;oldattr);&amp;nbsp;//&amp;nbsp;恢复终端属性
&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;return&amp;nbsp;ch;
}

__global__&amp;nbsp;void&amp;nbsp;what_is_my_id(unsigned&amp;nbsp;int&amp;nbsp;*&amp;nbsp;const&amp;nbsp;block,&amp;nbsp;unsigned&amp;nbsp;int&amp;nbsp;*&amp;nbsp;const&amp;nbsp;thread,
&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;unsigned&amp;nbsp;int&amp;nbsp;*&amp;nbsp;const&amp;nbsp;warp,&amp;nbsp;unsigned&amp;nbsp;int&amp;nbsp;*&amp;nbsp;const&amp;nbsp;calc_thread)
{
&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;//&amp;nbsp;thread&amp;nbsp;is&amp;nbsp;is&amp;nbsp;block&amp;nbsp;index&amp;nbsp;*&amp;nbsp;block&amp;nbsp;size&amp;nbsp;+&amp;nbsp;thread&amp;nbsp;offset&amp;nbsp;into&amp;nbsp;the&amp;nbsp;block&amp;nbsp;
&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;const&amp;nbsp;unsigned&amp;nbsp;int&amp;nbsp;thread_idx&amp;nbsp;=&amp;nbsp;(blockIdx.x&amp;nbsp;*&amp;nbsp;blockDim.x)&amp;nbsp;+&amp;nbsp;threadIdx.x;
&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;block[thread_idx]&amp;nbsp;=&amp;nbsp;blockIdx.x;
&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;thread[thread_idx]&amp;nbsp;=&amp;nbsp;threadIdx.x;

&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;//&amp;nbsp;calculate&amp;nbsp;warp&amp;nbsp;using&amp;nbsp;built&amp;nbsp;in&amp;nbsp;variable&amp;nbsp;warpSize
&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;warp[thread_idx]&amp;nbsp;=&amp;nbsp;threadIdx.x&amp;nbsp;/&amp;nbsp;warpSize;
&amp;nbsp;&amp;nbsp;&amp;nbsp;
&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;calc_thread[thread_idx]&amp;nbsp;=&amp;nbsp;thread_idx;
}

#define&amp;nbsp;ARRAY_SIZE&amp;nbsp;128
#define&amp;nbsp;ARRAY_SIZE_IN_BYTES&amp;nbsp;(sizeof(unsigned&amp;nbsp;int)&amp;nbsp;*&amp;nbsp;ARRAY_SIZE)

unsigned&amp;nbsp;int&amp;nbsp;cpu_block[ARRAY_SIZE];
unsigned&amp;nbsp;int&amp;nbsp;cpu_thread[ARRAY_SIZE];
unsigned&amp;nbsp;int&amp;nbsp;cpu_warp[ARRAY_SIZE];
unsigned&amp;nbsp;int&amp;nbsp;cpu_cal_thread[ARRAY_SIZE];


int&amp;nbsp;main(void)
{
&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;//&amp;nbsp;total&amp;nbsp;thread&amp;nbsp;count&amp;nbsp;=&amp;nbsp;2&amp;nbsp;*&amp;nbsp;62&amp;nbsp;=&amp;nbsp;128&amp;nbsp;
&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;const&amp;nbsp;unsigned&amp;nbsp;int&amp;nbsp;num_blocks&amp;nbsp;=&amp;nbsp;2;
&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;const&amp;nbsp;unsigned&amp;nbsp;int&amp;nbsp;num_threads&amp;nbsp;=&amp;nbsp;64;

&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;unsigned&amp;nbsp;int&amp;nbsp;*&amp;nbsp;gpu_block;
&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;unsigned&amp;nbsp;int&amp;nbsp;*&amp;nbsp;gpu_thread;
&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;unsigned&amp;nbsp;int&amp;nbsp;*&amp;nbsp;gpu_warp;
&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;unsigned&amp;nbsp;int&amp;nbsp;*&amp;nbsp;gpu_calc_thread;

&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;//&amp;nbsp;allocate&amp;nbsp;four&amp;nbsp;arrays&amp;nbsp;on&amp;nbsp;the&amp;nbsp;gpu&amp;nbsp;
&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;cudaMalloc((void&amp;nbsp;**)&amp;amp;gpu_block,&amp;nbsp;ARRAY_SIZE_IN_BYTES);
&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;cudaMalloc((void&amp;nbsp;**)&amp;amp;gpu_thread,&amp;nbsp;ARRAY_SIZE_IN_BYTES);
&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;cudaMalloc((void&amp;nbsp;**)&amp;amp;gpu_warp,&amp;nbsp;ARRAY_SIZE_IN_BYTES);
&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;cudaMalloc((void&amp;nbsp;**)&amp;amp;gpu_calc_thread,&amp;nbsp;ARRAY_SIZE_IN_BYTES);

&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;//&amp;nbsp;Execute&amp;nbsp;our&amp;nbsp;kernel&amp;nbsp;
&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;what_is_my_id&amp;lt;&amp;lt;&amp;lt;num_blocks,&amp;nbsp;num_threads&amp;gt;&amp;gt;&amp;gt;(gpu_block,&amp;nbsp;gpu_thread,&amp;nbsp;gpu_warp,&amp;nbsp;gpu_calc_thread);
&amp;nbsp;&amp;nbsp;&amp;nbsp;
&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;//&amp;nbsp;copy&amp;nbsp;back&amp;nbsp;the&amp;nbsp;gpu&amp;nbsp;results&amp;nbsp;to&amp;nbsp;the&amp;nbsp;cpu&amp;nbsp;
&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;cudaMemcpy(cpu_block,&amp;nbsp;gpu_block,&amp;nbsp;ARRAY_SIZE_IN_BYTES,&amp;nbsp;cudaMemcpyDeviceToHost);
&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;cudaMemcpy(cpu_thread,&amp;nbsp;gpu_thread,&amp;nbsp;ARRAY_SIZE_IN_BYTES,&amp;nbsp;cudaMemcpyDeviceToHost);
&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;cudaMemcpy(cpu_warp,&amp;nbsp;gpu_warp,&amp;nbsp;ARRAY_SIZE_IN_BYTES,&amp;nbsp;cudaMemcpyDeviceToHost);
&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;cudaMemcpy(cpu_cal_thread,&amp;nbsp;gpu_calc_thread,&amp;nbsp;ARRAY_SIZE_IN_BYTES,&amp;nbsp;cudaMemcpyDeviceToHost);

&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;//&amp;nbsp;free&amp;nbsp;the&amp;nbsp;arrays&amp;nbsp;on&amp;nbsp;the&amp;nbsp;gpu&amp;nbsp;as&amp;nbsp;now&amp;nbsp;we&amp;#39;re&amp;nbsp;done&amp;nbsp;with&amp;nbsp;them
&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;cudaFree(gpu_block);
&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;cudaFree(gpu_thread);
&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;cudaFree(gpu_warp);
&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;cudaFree(gpu_calc_thread);

&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;//&amp;nbsp;iterate&amp;nbsp;through&amp;nbsp;the&amp;nbsp;arrays&amp;nbsp;and&amp;nbsp;print&amp;nbsp;
&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;for(unsigned&amp;nbsp;int&amp;nbsp;i&amp;nbsp;=&amp;nbsp;0;&amp;nbsp;i&amp;nbsp;&amp;lt;&amp;nbsp;ARRAY_SIZE;&amp;nbsp;++i)&amp;nbsp;{
&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;printf(&amp;quot;calculate&amp;nbsp;thread:&amp;nbsp;%3u&amp;nbsp;-&amp;nbsp;block:&amp;nbsp;%2u&amp;nbsp;-&amp;nbsp;warp&amp;nbsp;%2u&amp;nbsp;-&amp;nbsp;thread&amp;nbsp;%3u\n&amp;quot;,&amp;nbsp;cpu_cal_thread[i],&amp;nbsp;cpu_block[i],&amp;nbsp;cpu_warp[i],&amp;nbsp;cpu_thread[i]);
&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;}
&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;getch();
&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;return&amp;nbsp;0;
}&lt;/pre&gt;</description><pubDate>Sun, 08 Sep 2024 23:13:30 +0800</pubDate></item><item><title>DETR端到端检测主流框架：DNIO论文解析</title><link>http://www.trizmed.cn/jscxd/18.html</link><description>&lt;p style=&quot;margin-bottom:0;text-indent:28px;text-align:justify;text-justify:inter-ideograph&quot;&gt;&lt;strong&gt;&lt;span style=&quot;font-family: &amp;#39;ＭＳ 明朝&amp;#39;;font-size: 14px&quot;&gt;&lt;span style=&quot;font-family:ＭＳ 明朝&quot;&gt;论文解读：&lt;/span&gt;&lt;span style=&quot;font-family:Cambria&quot;&gt;DETR with Improved DeNoising Anchor Boxes For End-to-End Object Detection&lt;/span&gt;&lt;/span&gt;&lt;/strong&gt;&lt;/p&gt;&lt;p style=&quot;margin-bottom:0;text-indent:28px;text-align:justify;text-justify:inter-ideograph&quot;&gt;&lt;strong&gt;&lt;span style=&quot;font-family: &amp;#39;ＭＳ 明朝&amp;#39;;font-size: 14px&quot;&gt;&amp;nbsp;&lt;/span&gt;&lt;/strong&gt;&lt;/p&gt;&lt;p style=&quot;margin-bottom: 0&quot;&gt;&lt;span style=&quot;font-family:&amp;#39;ＭＳ 明朝&amp;#39;;font-weight:bold;font-size:14px&quot;&gt;一、&lt;/span&gt;&lt;strong&gt;&lt;span style=&quot;font-family: &amp;#39;ＭＳ 明朝&amp;#39;;font-size: 14px&quot;&gt;摘要&lt;/span&gt;&lt;/strong&gt;&lt;/p&gt;&lt;p style=&quot;margin-bottom: 0;text-indent: 28px&quot;&gt;&lt;span style=&quot;;font-family:&amp;#39;ＭＳ 明朝&amp;#39;;font-size:14px&quot;&gt;&lt;span style=&quot;font-family:ＭＳ 明朝&quot;&gt;本文提出了&lt;/span&gt;&lt;span style=&quot;font-family:Cambria&quot;&gt;DINO&lt;/span&gt;&lt;span style=&quot;font-family:ＭＳ 明朝&quot;&gt;（&lt;/span&gt;&lt;/span&gt;&lt;span style=&quot;;font-family:&amp;#39;ＭＳ 明朝&amp;#39;;font-size:14px&quot;&gt;&lt;span style=&quot;font-family:Cambria&quot;&gt;DETR with Improved DeNoising Anchor Boxes&lt;/span&gt;&lt;/span&gt;&lt;span style=&quot;;font-family:&amp;#39;ＭＳ 明朝&amp;#39;;font-size:14px&quot;&gt;&lt;span style=&quot;font-family:ＭＳ 明朝&quot;&gt;），一种端到端目标检测器。&lt;/span&gt;&lt;span style=&quot;font-family:Cambria&quot;&gt;DINO&lt;/span&gt;&lt;span style=&quot;font-family:ＭＳ 明朝&quot;&gt;通过对去噪训练使用对比方法、混合查询选择方法进行锚点初始化、以及两次前向预测机制来改进性能和效率。&lt;/span&gt;&lt;span style=&quot;font-family:Cambria&quot;&gt;DINO&lt;/span&gt;&lt;span style=&quot;font-family:ＭＳ 明朝&quot;&gt;在&lt;/span&gt;&lt;span style=&quot;font-family:Cambria&quot;&gt;12&lt;/span&gt;&lt;span style=&quot;font-family:ＭＳ 明朝&quot;&gt;个周期内实现了&lt;/span&gt;&lt;span style=&quot;font-family:Cambria&quot;&gt;49.4 AP&lt;/span&gt;&lt;span style=&quot;font-family:ＭＳ 明朝&quot;&gt;，在&lt;/span&gt;&lt;span style=&quot;font-family:Cambria&quot;&gt;24&lt;/span&gt;&lt;span style=&quot;font-family:ＭＳ 明朝&quot;&gt;个周期内实现了&lt;/span&gt;&lt;span style=&quot;font-family:Cambria&quot;&gt;51.3 AP&lt;/span&gt;&lt;span style=&quot;font-family:ＭＳ 明朝&quot;&gt;（使用&lt;/span&gt;&lt;span style=&quot;font-family:Cambria&quot;&gt;ResNet-50&lt;/span&gt;&lt;span style=&quot;font-family:ＭＳ 明朝&quot;&gt;骨干网络和多尺度特征）。与之前最好的&lt;/span&gt;&lt;span style=&quot;font-family:Cambria&quot;&gt;DETR&lt;/span&gt;&lt;span style=&quot;font-family:ＭＳ 明朝&quot;&gt;模型&lt;/span&gt;&lt;span style=&quot;font-family:Cambria&quot;&gt;DN-DETR&lt;/span&gt;&lt;span style=&quot;font-family:ＭＳ 明朝&quot;&gt;相比，分别有&lt;/span&gt;&lt;span style=&quot;font-family:Cambria&quot;&gt;+6.0 AP&lt;/span&gt;&lt;span style=&quot;font-family:ＭＳ 明朝&quot;&gt;和&lt;/span&gt;&lt;span style=&quot;font-family:Cambria&quot;&gt;+2.7 AP&lt;/span&gt;&lt;span style=&quot;font-family:ＭＳ 明朝&quot;&gt;的显著提升。&lt;/span&gt;&lt;span style=&quot;font-family:Cambria&quot;&gt;DINO&lt;/span&gt;&lt;span style=&quot;font-family:ＭＳ 明朝&quot;&gt;在模型大小和数据规模上表现出良好的可扩展性。在&lt;/span&gt;&lt;span style=&quot;font-family:Cambria&quot;&gt;Objects365&lt;/span&gt;&lt;span style=&quot;font-family:ＭＳ 明朝&quot;&gt;数据集上使用&lt;/span&gt;&lt;span style=&quot;font-family:Cambria&quot;&gt;SwinL&lt;/span&gt;&lt;span style=&quot;font-family:ＭＳ 明朝&quot;&gt;骨干网络预训练后，&lt;/span&gt;&lt;span style=&quot;font-family:Cambria&quot;&gt;DINO&lt;/span&gt;&lt;span style=&quot;font-family:ＭＳ 明朝&quot;&gt;在&lt;/span&gt;&lt;span style=&quot;font-family:Cambria&quot;&gt;COCO val2017&lt;/span&gt;&lt;span style=&quot;font-family:ＭＳ 明朝&quot;&gt;和&lt;/span&gt;&lt;span style=&quot;font-family:Cambria&quot;&gt;test-dev&lt;/span&gt;&lt;span style=&quot;font-family:ＭＳ 明朝&quot;&gt;上均取得了最佳结果。与其他排行榜上的模型相比，&lt;/span&gt;&lt;span style=&quot;font-family:Cambria&quot;&gt;DINO&lt;/span&gt;&lt;span style=&quot;font-family:ＭＳ 明朝&quot;&gt;显著减少了模型大小和预训练数据量，同时取得了更好的结果。&lt;/span&gt;&lt;/span&gt;&lt;/p&gt;&lt;p style=&quot;margin-bottom: 0;text-indent: 28px&quot;&gt;&lt;span style=&quot;;font-family:&amp;#39;ＭＳ 明朝&amp;#39;;font-size:14px&quot;&gt;&lt;span style=&quot;font-family:ＭＳ 明朝&quot;&gt;&lt;br/&gt;&lt;/span&gt;&lt;/span&gt;&lt;/p&gt;&lt;p style=&quot;margin-bottom: 0px; text-align: left; text-indent: 0em;&quot;&gt;&lt;span style=&quot;font-family: &amp;quot;ＭＳ 明朝&amp;quot;; font-weight: bold;&quot;&gt;二、&lt;/span&gt;&lt;strong&gt;&lt;span style=&quot;font-family: &amp;quot;ＭＳ 明朝&amp;quot;;&quot;&gt;论文要解决的问题&lt;/span&gt;&lt;/strong&gt;&lt;/p&gt;&lt;p style=&quot;margin-bottom: 0;text-indent: 28px&quot;&gt;&lt;span style=&quot;;font-family:&amp;#39;ＭＳ 明朝&amp;#39;;font-size:14px&quot;&gt;该论文的主要&lt;/span&gt;&lt;span style=&quot;;font-family:&amp;#39;ＭＳ 明朝&amp;#39;;font-size:14px&quot;&gt;想解决的&lt;/span&gt;&lt;span style=&quot;;font-family:&amp;#39;ＭＳ 明朝&amp;#39;;font-size:14px&quot;&gt;&lt;span style=&quot;font-family:ＭＳ 明朝&quot;&gt;问题是现有&lt;/span&gt;&lt;span style=&quot;font-family:Cambria&quot;&gt;DETR&lt;/span&gt;&lt;span style=&quot;font-family:ＭＳ 明朝&quot;&gt;（&lt;/span&gt;&lt;span style=&quot;font-family:Cambria&quot;&gt;DEtection TRansformer&lt;/span&gt;&lt;span style=&quot;font-family:ＭＳ 明朝&quot;&gt;）模型在目标检测中的收敛速度慢且性能不理想。虽然&lt;/span&gt;&lt;span style=&quot;font-family:Cambria&quot;&gt;DETR&lt;/span&gt;&lt;span style=&quot;font-family:ＭＳ 明朝&quot;&gt;消除了诸如锚点生成和非极大值抑制等手工设计的组件，但其训练速度较慢且查询的意义不够明确，导致其性能落后于传统检测器。此外，现有的&lt;/span&gt;&lt;span style=&quot;font-family:Cambria&quot;&gt;DETR&lt;/span&gt;&lt;span style=&quot;font-family:ＭＳ 明朝&quot;&gt;类模型在应用于更大的骨干网络和数据集时，其可扩展性尚未得到充分研究。&lt;/span&gt;&lt;/span&gt;&lt;/p&gt;&lt;p style=&quot;margin-bottom: 0;text-indent: 29px&quot;&gt;&lt;span style=&quot;;font-family:Cambria;font-size:15px&quot;&gt;&lt;br/&gt;&lt;/span&gt;&lt;/p&gt;&lt;p style=&quot;text-align:center&quot;&gt;&lt;img class=&quot;ue-image&quot; src=&quot;http://www.trizmed.cn/zb_users/upload/2024/09/202409081725804300404181.png&quot; title=&quot;图片1.png&quot; alt=&quot;图片1.png&quot; width=&quot;699&quot; height=&quot;222&quot; style=&quot;width: 699px; height: 222px;&quot;/&gt;&lt;/p&gt;&lt;p style=&quot;text-align: center;&quot;&gt;&lt;span style=&quot;font-size: 10px; text-indent: 28px; font-family: &amp;quot;ＭＳ 明朝&amp;quot;;&quot;&gt;图1&lt;/span&gt;&lt;span style=&quot;font-family: &amp;quot;Times New Roman&amp;quot;; font-size: 10px; text-indent: 28px;&quot;&gt;&amp;nbsp;DINO&lt;/span&gt;&lt;span style=&quot;font-size: 10px; text-indent: 28px; font-family: &amp;quot;ＭＳ 明朝&amp;quot;;&quot;&gt;系统结构图&lt;/span&gt;&lt;/p&gt;&lt;p style=&quot;margin-bottom: 0;margin-left: 28px&quot;&gt;&lt;span style=&quot;font-family:&amp;#39;ＭＳ 明朝&amp;#39;;font-weight:bold;font-size:14px&quot;&gt;三、&lt;/span&gt;&lt;strong&gt;&lt;span style=&quot;font-family: &amp;#39;ＭＳ 明朝&amp;#39;;font-size: 14px&quot;&gt;解决方案&lt;/span&gt;&lt;/strong&gt;&lt;/p&gt;&lt;p style=&quot;text-indent:28px;text-autospace:ideograph-numeric&quot;&gt;&lt;span style=&quot;;font-family:&amp;#39;ＭＳ 明朝&amp;#39;;font-size:14px&quot;&gt;为&lt;/span&gt;&lt;span style=&quot;;font-family:&amp;#39;ＭＳ 明朝&amp;#39;;font-size:14px&quot;&gt;了解决这些问题，论文提出了一种新模型，引入了几个关键创新：&lt;/span&gt;&lt;span style=&quot;;font-family:&amp;#39;ＭＳ 明朝&amp;#39;;font-size:14px&quot;&gt;&lt;br/&gt;&lt;/span&gt;&lt;span style=&quot;;font-family:&amp;#39;ＭＳ 明朝&amp;#39;;font-size:14px&quot;&gt;&lt;span style=&quot;font-family:Cambria&quot;&gt;1. &lt;/span&gt;&lt;span style=&quot;font-family:ＭＳ 明朝&quot;&gt;对比去噪训练（&lt;/span&gt;&lt;span style=&quot;font-family:Cambria&quot;&gt;Contrastive Denoising Training&lt;/span&gt;&lt;span style=&quot;font-family:ＭＳ 明朝&quot;&gt;）：传统的&lt;/span&gt;&lt;span style=&quot;font-family:Cambria&quot;&gt;DETR&lt;/span&gt;&lt;span style=&quot;font-family:ＭＳ 明朝&quot;&gt;模型在训练过程中容易产生不稳定的双向匹配问题。&lt;/span&gt;&lt;span style=&quot;font-family:Cambria&quot;&gt;DINO&lt;/span&gt;&lt;span style=&quot;font-family:ＭＳ 明朝&quot;&gt;通过引入对比去噪训练来解决这个问题，即同时加入正样本和负样本，增强模型的学习能力。具体而言，&lt;/span&gt;&lt;span style=&quot;font-family:Cambria&quot;&gt;DINO&lt;/span&gt;&lt;span style=&quot;font-family:ＭＳ 明朝&quot;&gt;在每个真实框中添加两种不同噪声的版本，其中噪声较小的版本被标记为正样本，较大的噪声版本被标记为负样本。这种方法有助于模型区分不同的目标，避免重复检测相同的对象，提高检测的准确性和鲁棒性。&lt;/span&gt;&lt;/span&gt;&lt;/p&gt;&lt;p style=&quot;text-align: center; text-indent: 2em;&quot;&gt;&lt;img class=&quot;ue-image&quot; src=&quot;http://www.trizmed.cn/zb_users/upload/2024/09/202409081725804325144143.png&quot; title=&quot;图片2.png&quot; alt=&quot;图片2.png&quot; width=&quot;464&quot; height=&quot;244&quot; style=&quot;width: 464px; height: 244px;&quot;/&gt;&lt;span style=&quot;;font-family:&amp;#39;ＭＳ 明朝&amp;#39;;font-size:14px&quot;&gt;&lt;br/&gt;&lt;/span&gt;&lt;span style=&quot;;font-family:&amp;#39;ＭＳ 明朝&amp;#39;;font-size:14px&quot;&gt;&lt;span style=&quot;font-family:Cambria&quot;&gt;2. &lt;/span&gt;&lt;span style=&quot;font-family:ＭＳ 明朝&quot;&gt;混合查询选择方法（&lt;/span&gt;&lt;span style=&quot;font-family:Cambria&quot;&gt;Mixed Query Selection Method&lt;/span&gt;&lt;span style=&quot;font-family:ＭＳ 明朝&quot;&gt;）：&lt;/span&gt;&lt;span style=&quot;font-family:Cambria&quot;&gt;DETR&lt;/span&gt;&lt;span style=&quot;font-family:ＭＳ 明朝&quot;&gt;模型中的查询是静态的，这导致了在不同图像的推理中缺乏灵活性。&lt;/span&gt;&lt;span style=&quot;font-family:Cambria&quot;&gt;DINO&lt;/span&gt;&lt;span style=&quot;font-family:ＭＳ 明朝&quot;&gt;采用混合查询选择方法，通过利用来自编码器输出的位置信息初始化锚框，同时保持内容查询的可学习性，从而改进了解码器中的查询初始化过程。这种方法使得模型能够更有效地结合空间和内容信息，提高检测精度。&lt;/span&gt;&lt;/span&gt;&lt;/p&gt;&lt;p style=&quot;text-indent: 0em; text-align: center;&quot;&gt;&lt;img class=&quot;ue-image&quot; src=&quot;http://www.trizmed.cn/zb_users/upload/2024/09/202409081725804351528356.png&quot; title=&quot;图片3.png&quot; alt=&quot;图片3.png&quot; width=&quot;605&quot; height=&quot;257&quot; style=&quot;width: 605px; height: 257px;&quot;/&gt;&lt;span style=&quot;;font-family:&amp;#39;ＭＳ 明朝&amp;#39;;font-size:14px&quot;&gt;&lt;br/&gt;&lt;/span&gt;&lt;span style=&quot;;font-family:&amp;#39;ＭＳ 明朝&amp;#39;;font-size:14px&quot;&gt;&lt;span style=&quot;font-family:Cambria&quot;&gt;3. &lt;/span&gt;&lt;span style=&quot;font-family:ＭＳ 明朝&quot;&gt;两次前向预测机制（&lt;/span&gt;&lt;span style=&quot;font-family:Cambria&quot;&gt;Look Forward Twice Scheme&lt;/span&gt;&lt;span style=&quot;font-family:ＭＳ 明朝&quot;&gt;）：&lt;/span&gt;&lt;span style=&quot;font-family:Cambria&quot;&gt;DINO&lt;/span&gt;&lt;span style=&quot;font-family:ＭＳ 明朝&quot;&gt;提出了一种新的预测机制，通过使用后层的精细信息来修正早期层的参数优化。在传统方法中，每一层的参数仅基于当前层的损失进行更新，而&lt;/span&gt;&lt;span style=&quot;font-family:Cambria&quot;&gt;DINO&lt;/span&gt;&lt;span style=&quot;font-family:ＭＳ 明朝&quot;&gt;通过引入“向前看两次”的方法，将后续层的信息反馈到前一层，从而实现更好的参数优化。这种方法能够更好地利用深层信息来改进早期预测，提高模型的整体性能。&lt;/span&gt;&lt;/span&gt;&lt;/p&gt;&lt;p style=&quot;text-indent: 28px&quot;&gt;&lt;span style=&quot;;font-family:&amp;#39;ＭＳ 明朝&amp;#39;;font-size:14px&quot;&gt;&lt;img class=&quot;ue-image&quot; src=&quot;http://www.trizmed.cn/zb_users/upload/2024/09/202409081725804384157373.png&quot; title=&quot;图片4.png&quot; alt=&quot;图片4.png&quot; width=&quot;646&quot; height=&quot;216&quot; style=&quot;width: 646px; height: 216px;&quot;/&gt;&amp;nbsp;&lt;/span&gt;&lt;/p&gt;&lt;p style=&quot;margin-bottom:0;text-autospace:ideograph-numeric&quot;&gt;&lt;strong&gt;&lt;span style=&quot;font-family: &amp;#39;ＭＳ 明朝&amp;#39;;font-size: 14px&quot;&gt;四、实验分析&lt;/span&gt;&lt;/strong&gt;&lt;/p&gt;&lt;p style=&quot;margin-bottom:0;text-autospace:ideograph-numeric&quot;&gt;&lt;span style=&quot;;font-family:&amp;#39;ＭＳ 明朝&amp;#39;;font-size:14px&quot;&gt;论文进行了&lt;/span&gt;&lt;span style=&quot;;font-family:&amp;#39;ＭＳ 明朝&amp;#39;;font-size:14px&quot;&gt;大量&lt;/span&gt;&lt;span style=&quot;;font-family:&amp;#39;ＭＳ 明朝&amp;#39;;font-size:14px&quot;&gt;&lt;span style=&quot;font-family:ＭＳ 明朝&quot;&gt;的实验来验证&lt;/span&gt;&lt;span style=&quot;font-family:Cambria&quot;&gt;DINO&lt;/span&gt;&lt;span style=&quot;font-family:ＭＳ 明朝&quot;&gt;的有效性。关键结果包括：&lt;/span&gt;&lt;/span&gt;&lt;span style=&quot;;font-family:&amp;#39;ＭＳ 明朝&amp;#39;;font-size:14px&quot;&gt;&lt;br/&gt;&lt;/span&gt;&lt;span style=&quot;;font-family:&amp;#39;ＭＳ 明朝&amp;#39;;font-size:14px&quot;&gt;&lt;span style=&quot;font-family:Cambria&quot;&gt;1. DINO&lt;/span&gt;&lt;span style=&quot;font-family:ＭＳ 明朝&quot;&gt;在&lt;/span&gt;&lt;span style=&quot;font-family:Cambria&quot;&gt;12&lt;/span&gt;&lt;span style=&quot;font-family:ＭＳ 明朝&quot;&gt;个周期内达到&lt;/span&gt;&lt;span style=&quot;font-family:Cambria&quot;&gt;49.4 AP&lt;/span&gt;&lt;span style=&quot;font-family:ＭＳ 明朝&quot;&gt;，在&lt;/span&gt;&lt;span style=&quot;font-family:Cambria&quot;&gt;24&lt;/span&gt;&lt;span style=&quot;font-family:ＭＳ 明朝&quot;&gt;个周期内达到&lt;/span&gt;&lt;span style=&quot;font-family:Cambria&quot;&gt;51.3 AP&lt;/span&gt;&lt;span style=&quot;font-family:ＭＳ 明朝&quot;&gt;，使用&lt;/span&gt;&lt;span style=&quot;font-family:Cambria&quot;&gt;ResNet-50&lt;/span&gt;&lt;span style=&quot;font-family:ＭＳ 明朝&quot;&gt;骨干网络，分别显著超过之前的&lt;/span&gt;&lt;span style=&quot;font-family:Cambria&quot;&gt;DETR&lt;/span&gt;&lt;span style=&quot;font-family:ＭＳ 明朝&quot;&gt;类模型&lt;/span&gt;&lt;span style=&quot;font-family:Cambria&quot;&gt;+6.0 AP&lt;/span&gt;&lt;span style=&quot;font-family:ＭＳ 明朝&quot;&gt;和&lt;/span&gt;&lt;span style=&quot;font-family:Cambria&quot;&gt;+2.7 AP&lt;/span&gt;&lt;span style=&quot;font-family:ＭＳ 明朝&quot;&gt;。&lt;/span&gt;&lt;/span&gt;&lt;/p&gt;&lt;p style=&quot;margin-bottom: 0px; text-align: center;&quot;&gt;&lt;span style=&quot;;font-family:&amp;#39;ＭＳ 明朝&amp;#39;;font-size:14px&quot;&gt;&lt;img class=&quot;ue-image&quot; src=&quot;http://www.trizmed.cn/zb_users/upload/2024/09/202409081725804407300789.png&quot; title=&quot;图片5.png&quot; alt=&quot;图片5.png&quot; width=&quot;639&quot; height=&quot;182&quot; style=&quot;width: 639px; height: 182px;&quot;/&gt;&lt;/span&gt;&lt;span style=&quot;;font-family:&amp;#39;ＭＳ 明朝&amp;#39;;font-size:14px&quot;&gt;&lt;span style=&quot;font-family:Cambria&quot;&gt;2. &lt;/span&gt;&lt;span style=&quot;font-family:ＭＳ 明朝&quot;&gt;当在&lt;/span&gt;&lt;span style=&quot;font-family:Cambria&quot;&gt;Objects365&lt;/span&gt;&lt;span style=&quot;font-family:ＭＳ 明朝&quot;&gt;数据集上使用&lt;/span&gt;&lt;span style=&quot;font-family:Cambria&quot;&gt;SwinL&lt;/span&gt;&lt;span style=&quot;font-family:ＭＳ 明朝&quot;&gt;骨干网络进行预训练后，&lt;/span&gt;&lt;span style=&quot;font-family:Cambria&quot;&gt;DINO&lt;/span&gt;&lt;span style=&quot;font-family:ＭＳ 明朝&quot;&gt;在&lt;/span&gt;&lt;span style=&quot;font-family:Cambria&quot;&gt;COCO val2017&lt;/span&gt;&lt;span style=&quot;font-family:ＭＳ 明朝&quot;&gt;上达到&lt;/span&gt;&lt;span style=&quot;font-family:Cambria&quot;&gt;63.2 AP&lt;/span&gt;&lt;span style=&quot;font-family:ＭＳ 明朝&quot;&gt;，在&lt;/span&gt;&lt;span style=&quot;font-family:Cambria&quot;&gt;test-dev&lt;/span&gt;&lt;span style=&quot;font-family:ＭＳ 明朝&quot;&gt;上达到&lt;/span&gt;&lt;span style=&quot;font-family:Cambria&quot;&gt;63.3 AP&lt;/span&gt;&lt;span style=&quot;font-family:ＭＳ 明朝&quot;&gt;，创下新的最先进记录（&lt;/span&gt;&lt;span style=&quot;font-family:Cambria&quot;&gt;SOTA&lt;/span&gt;&lt;span style=&quot;font-family:ＭＳ 明朝&quot;&gt;）。&lt;/span&gt;&lt;/span&gt;&lt;/p&gt;&lt;p style=&quot;margin-bottom: 0px; text-align: center;&quot;&gt;&lt;img class=&quot;ue-image&quot; src=&quot;http://www.trizmed.cn/zb_users/upload/2024/09/202409081725804419132804.png&quot; title=&quot;图片6.png&quot; alt=&quot;图片6.png&quot; width=&quot;459&quot; height=&quot;196&quot; style=&quot;width: 459px; height: 196px;&quot;/&gt;&lt;/p&gt;&lt;p style=&quot;margin-bottom: 0px;&quot;&gt;&lt;span style=&quot;;font-family:&amp;#39;ＭＳ 明朝&amp;#39;;font-size:14px&quot;&gt;&lt;span style=&quot;font-family:Cambria&quot;&gt;3. DINO&lt;/span&gt;&lt;span style=&quot;font-family:ＭＳ 明朝&quot;&gt;显示出显著的可扩展性，在保持优异性能的同时，减少了模型大小和预训练数据量。&lt;/span&gt;&lt;/span&gt;&lt;span style=&quot;;font-family:&amp;#39;ＭＳ 明朝&amp;#39;;font-size:14px&quot;&gt;&lt;br/&gt;&lt;img class=&quot;ue-image&quot; src=&quot;http://www.trizmed.cn/zb_users/upload/2024/09/202409081725804449380353.png&quot; title=&quot;图片7.png&quot; alt=&quot;图片7.png&quot; width=&quot;704&quot; height=&quot;159&quot; style=&quot;width: 704px; height: 159px;&quot;/&gt;&lt;/span&gt;&lt;/p&gt;&lt;p style=&quot;margin-bottom:0;text-autospace:ideograph-numeric&quot;&gt;&lt;strong&gt;&lt;span style=&quot;font-family: &amp;#39;ＭＳ 明朝&amp;#39;;font-size: 14px&quot;&gt;五、论文总结&lt;/span&gt;&lt;/strong&gt;&lt;/p&gt;&lt;p style=&quot;margin-bottom:0;text-indent:28px;text-autospace:ideograph-numeric&quot;&gt;&lt;span style=&quot;;font-family:&amp;#39;ＭＳ 明朝&amp;#39;;font-size:14px&quot;&gt;&lt;span style=&quot;font-family:ＭＳ 明朝&quot;&gt;通过引入创新的对比去噪训练、混合查询选择和两次前向预测机制，&lt;/span&gt;&lt;span style=&quot;font-family:Cambria&quot;&gt;DINO&lt;/span&gt;&lt;span style=&quot;font-family:ＭＳ 明朝&quot;&gt;成功解决了现有&lt;/span&gt;&lt;span style=&quot;font-family:Cambria&quot;&gt;DETR&lt;/span&gt;&lt;span style=&quot;font-family:ＭＳ 明朝&quot;&gt;模型在目标检测任务中的核心局限性。实验结果表明，&lt;/span&gt;&lt;span style=&quot;font-family:Cambria&quot;&gt;DINO&lt;/span&gt;&lt;span style=&quot;font-family:ＭＳ 明朝&quot;&gt;不仅在收敛速度和检测性能上显著优于之前的模型，还展示了在更大规模数据集和更复杂模型架构上的良好扩展性。这一成果将&lt;/span&gt;&lt;span style=&quot;font-family:Cambria&quot;&gt;DETR&lt;/span&gt;&lt;span style=&quot;font-family:ＭＳ 明朝&quot;&gt;类模型确立为一种主流的检测框架，不仅仅因为其新颖的端到端检测优化方式，更因为其在性能上的卓越表现。&lt;/span&gt;&lt;/span&gt;&lt;/p&gt;&lt;p&gt;&lt;br/&gt;&lt;/p&gt;</description><pubDate>Sun, 08 Sep 2024 22:01:55 +0800</pubDate></item><item><title>PowerInfer-2: Fast Large Language Model Inference on a Smartphone论文解读</title><link>http://www.trizmed.cn/jscxd/17.html</link><description>&lt;p class=&quot;ue-upload&quot; style=&quot;line-height: 16px;&quot;&gt;&lt;img style=&quot;vertical-align: middle; margin-right: 2px;&quot; src=&quot;http://www.trizmed.cn/zb_system/image/filetype/pdf.png&quot;/&gt;&lt;a style=&quot;font-size:12px; color:#0066cc;&quot; href=&quot;http://www.trizmed.cn/zb_users/upload/2024/09/202409021725283045258776.pdf&quot; title=&quot;PowerInfer-2 Fast Large Language Model Inference on a Smartphone.pdf&quot; target=&quot;_blank&quot;&gt;PowerInfer-2 Fast Large Language Model Inference on a Smartphone.pdf&lt;/a&gt; &amp;nbsp; &amp;nbsp;
 &amp;nbsp;&lt;iframe src=&quot;http://www.trizmed.cn/zb_users/upload/2024/09/202409021725283045258776.pdf&quot; height=&quot;600px&quot; width=&quot;100%&quot;&gt;&lt;/iframe&gt;&lt;/p&gt;&lt;p&gt;&lt;br/&gt;&lt;/p&gt;</description><pubDate>Mon, 02 Sep 2024 21:09:43 +0800</pubDate></item><item><title>同华为同学共同完成的技术方案获得美国专利授权-一种xx图像处理方法</title><link>http://www.trizmed.cn/lwjd/16.html</link><description>&lt;p&gt;&lt;img class=&quot;ue-image&quot; src=&quot;http://www.trizmed.cn/zb_users/upload/2024/09/202409011725205251363181.png&quot; title=&quot;shuailin-us.png&quot; alt=&quot;shuailin-us.png&quot;&gt;&lt;/p&gt;</description><pubDate>Sun, 01 Sep 2024 23:37:20 +0800</pubDate></item><item><title>同快手同学共同完成的技术方案授权-一种音量xx方法、装置、电子设备及存储介质</title><link>http://www.trizmed.cn/lwjd/15.html</link><description>&lt;p&gt;&lt;img class=&quot;ue-image&quot; src=&quot;http://www.trizmed.cn/zb_users/upload/2024/09/202409011725204986849528.png&quot; title=&quot;huangjinwen0.png&quot; alt=&quot;huangjinwen0.png&quot;&gt;&lt;/p&gt;</description><pubDate>Sun, 01 Sep 2024 23:34:09 +0800</pubDate></item><item><title>同快手同学共同完成的技术方案授权-一种xx展示方法、装置、电子设备及存储介质</title><link>http://www.trizmed.cn/lwjd/14.html</link><description>&lt;p&gt;&lt;img class=&quot;ue-image&quot; src=&quot;http://www.trizmed.cn/zb_users/upload/2024/09/202409011725204657263655.png&quot; title=&quot;zhaozhibiao.png&quot; alt=&quot;zhaozhibiao.png&quot;&gt;&lt;/p&gt;</description><pubDate>Sun, 01 Sep 2024 23:26:00 +0800</pubDate></item><item><title>论文：Mask DINO 解析</title><link>http://www.trizmed.cn/jscxd/12.html</link><description>&lt;p style=&quot;-w t, &amp;#39;Helvetica Neue&amp;#39;, &amp;#39;Hiragino Sans GB&amp;#39;, &amp;#39;Microsoft YaHei UI&amp;#39;, &amp;#39;Microsoft YaHei&amp;#39;, Arial, sans-serif;font-size: 17px;letter-spacing: 0.544px;text-align: justify;text-wrap: wrap;background-color: rgb(255, 255, 255);visibility: visible;box-sizing: border-box !important;overflow-wrap: break-word !important&quot;&gt;&lt;strong style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important;visibility: visible&quot;&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important;font-family: 宋体;font-size: 14px;visibility: visible&quot;&gt;一、&lt;/span&gt;&lt;/strong&gt;&lt;strong style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important;visibility: visible&quot;&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important;font-family: 宋体;font-size: 14px;visibility: visible&quot;&gt;摘要&lt;/span&gt;&lt;/strong&gt;&lt;strong style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important;visibility: visible&quot;&gt;&lt;/strong&gt;&lt;/p&gt;&lt;p style=&quot;-w t, &amp;#39;Helvetica Neue&amp;#39;, &amp;#39;Hiragino Sans GB&amp;#39;, &amp;#39;Microsoft YaHei UI&amp;#39;, &amp;#39;Microsoft YaHei&amp;#39;, Arial, sans-serif;font-size: 17px;letter-spacing: 0.544px;text-align: justify;text-wrap: wrap;background-color: rgb(255, 255, 255);visibility: visible;box-sizing: border-box !important;overflow-wrap: break-word !important&quot;&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important;font-family: 宋体;font-size: 14px;visibility: visible&quot;&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;visibility: visible;box-sizing: border-box !important;overflow-wrap: break-word !important&quot;&gt;&amp;nbsp; 本文介绍了一种名为&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important;font-family: Calibri;visibility: visible&quot;&gt;Mask DINO&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;visibility: visible;box-sizing: border-box !important;overflow-wrap: break-word !important&quot;&gt;的创新框架，旨在统一目标检测与图像分割任务。通过在&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important;font-family: Calibri;visibility: visible&quot;&gt;DINO&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;visibility: visible;box-sizing: border-box !important;overflow-wrap: break-word !important&quot;&gt;（&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important;font-family: Calibri;visibility: visible&quot;&gt;DETR with Improved Denoising Anchor Boxes&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;visibility: visible;box-sizing: border-box !important;overflow-wrap: break-word !important&quot;&gt;）模型的基础上引入一个掩码预测分支，&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important;font-family: Calibri;visibility: visible&quot;&gt;Mask DINO&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;visibility: visible;box-sizing: border-box !important;overflow-wrap: break-word !important&quot;&gt;能够支持包括实例分割、全景分割和语义分割在内的所有图像分割任务。该框架利用&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important;font-family: Calibri;visibility: visible&quot;&gt;DINO&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;visibility: visible;box-sizing: border-box !important;overflow-wrap: break-word !important&quot;&gt;的查询嵌入，通过与高分辨率像素嵌入图的点积操作来预测一组二值掩码。&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important;font-family: Calibri;visibility: visible&quot;&gt;Mask DINO&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;visibility: visible;box-sizing: border-box !important;overflow-wrap: break-word !important&quot;&gt;的关键组件被扩展以适应分割任务，并通过共享架构和训练过程实现。该设计简洁高效，具有良好的扩展性，并能从大规模联合检测和分割数据集中获益。实验结果表明，&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important;font-family: Calibri;visibility: visible&quot;&gt;Mask DINO&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;visibility: visible;box-sizing: border-box !important;overflow-wrap: break-word !important&quot;&gt;在各种条件下均显著优于现有的专门化分割方法，无论是在基于&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important;font-family: Calibri;visibility: visible&quot;&gt;ResNet-50&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;visibility: visible;box-sizing: border-box !important;overflow-wrap: break-word !important&quot;&gt;的主干网络上，还是在使用&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important;font-family: Calibri;visibility: visible&quot;&gt;SwinL&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;visibility: visible;box-sizing: border-box !important;overflow-wrap: break-word !important&quot;&gt;主干网络的预训练模型中。特别是，在&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important;font-family: Calibri;visibility: visible&quot;&gt;COCO&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;visibility: visible;box-sizing: border-box !important;overflow-wrap: break-word !important&quot;&gt;数据集上的实例分割（&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important;font-family: Calibri;visibility: visible&quot;&gt;54.5 AP&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;visibility: visible;box-sizing: border-box !important;overflow-wrap: break-word !important&quot;&gt;）、全景分割（&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important;font-family: Calibri;visibility: visible&quot;&gt;59.4 PQ&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;visibility: visible;box-sizing: border-box !important;overflow-wrap: break-word !important&quot;&gt;）和&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important;font-family: Calibri;visibility: visible&quot;&gt;ADE20K&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;visibility: visible;box-sizing: border-box !important;overflow-wrap: break-word !important&quot;&gt;数据集上的语义分割（&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important;font-family: Calibri;visibility: visible&quot;&gt;60.8 mIoU&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;visibility: visible;box-sizing: border-box !important;overflow-wrap: break-word !important&quot;&gt;）方面，&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important;font-family: Calibri;visibility: visible&quot;&gt;Mask DINO&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;visibility: visible;box-sizing: border-box !important;overflow-wrap: break-word !important&quot;&gt;取得了当前最佳的结果。代码已开源：https://github.com/IDEA-Research/MaskDINO&lt;/span&gt;&lt;/span&gt;&lt;/p&gt;&lt;p style=&quot;-w t, &amp;#39;H g: border-box !important;overflow-wrap: break-word !important&quot;&gt;&lt;img class=&quot;ue-image&quot; src=&quot;http://www.trizmed.cn/zb_users/upload/2024/08/202408311725090961233673.png&quot; title=&quot;framework.png&quot; alt=&quot;framework.png&quot;/&gt;&lt;/p&gt;&lt;p style=&quot;-w t, &amp;quot;Helvetica Neue&amp;quot;, &amp;quot;Hiragino Sans GB&amp;quot;, &amp;quot;Microsoft YaHei UI&amp;quot;, &amp;quot;Microsoft YaHei&amp;quot;, Arial, sans-serif; font-size: 17px; letter-spacing: 0.544px; text-wrap: wrap; background-color: rgb(255, 255, 255); visibility: visible; box-sizing: border-box !important; overflow-wrap: break-word !important; text-align: center;&quot;&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important;font-size: 12px&quot;&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;letter-spacing: 0.034em;font-family: 宋体;box-sizing: border-box !important;overflow-wrap: break-word !important&quot;&gt;图&lt;/span&gt;&lt;span style=&quot;-w t;overflow-wrap: break-word !important&quot;&gt;1 Mask DINO&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;letter-spacing: 0.034em;font-family: 宋体;box-sizing: border-box !important;overflow-wrap: break-word !important&quot;&gt;系统结构图&lt;/span&gt;&lt;/span&gt;&lt;/p&gt;&lt;p style=&quot;-w t, &amp;#39;Helvetica Neue&amp;#39;, &amp;#39;Hiragino Sans GB&amp;#39;, &amp;#39;Microsoft YaHei UI&amp;#39;, &amp;#39;Microsoft YaHei&amp;#39;, Arial, sans-serif;font-size: 17px;letter-spacing: 0.544px;text-align: justify;text-wrap: wrap;background-color: rgb(255, 255, 255);box-sizing: border-box !important;overflow-wrap: break-word !important&quot;&gt;&lt;strong style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important&quot;&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important;font-family: 宋体;font-size: 14px&quot;&gt;二、&lt;/span&gt;&lt;/strong&gt;&lt;strong style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important&quot;&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important;font-family: 宋体;font-size: 14px&quot;&gt;论文核心创新点&lt;/span&gt;&lt;/strong&gt;&lt;/p&gt;&lt;p style=&quot;-w t, &amp;#39;Helvetica Neue&amp;#39;, &amp;#39;Hiragino Sans GB&amp;#39;, &amp;#39;Microsoft YaHei UI&amp;#39;, &amp;#39;Microsoft YaHei&amp;#39;, Arial, sans-serif;font-size: 17px;letter-spacing: 0.544px;text-align: justify;text-wrap: wrap;background-color: rgb(255, 255, 255);box-sizing: border-box !important;overflow-wrap: break-word !important&quot;&gt;&lt;span style=&quot;-w t-family: Calibri&quot;&gt;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;1.&amp;nbsp;&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important;font-family: 宋体;font-size: 14px;letter-spacing: 0.034em&quot;&gt;提出统一的目标检测和分割框架：&lt;/span&gt;&lt;span style=&quot;-w t-family: Calibri&quot;&gt;Mask DINO&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important;font-family: 宋体;font-size: 14px;letter-spacing: 0.034em&quot;&gt;通过在&lt;/span&gt;&lt;span style=&quot;-w t-family: Calibri&quot;&gt;DINO&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important;font-family: 宋体;font-size: 14px;letter-spacing: 0.034em&quot;&gt;模型上增加一个掩码预测分支，实现了对检测和多种分割任务的统一处理，简化了算法开发，同时在多个任务上提升了性能。&lt;/span&gt;&lt;/p&gt;&lt;p style=&quot;-w t, &amp;#39;Helvetica Neue&amp;#39;, &amp;#39;Hiragino Sans GB&amp;#39;, &amp;#39;Microsoft YaHei UI&amp;#39;, &amp;#39;Microsoft YaHei&amp;#39;, Arial, sans-serif;font-size: 17px;letter-spacing: 0.544px;text-align: justify;text-wrap: wrap;background-color: rgb(255, 255, 255);box-sizing: border-box !important;overflow-wrap: break-word !important&quot;&gt;&lt;span style=&quot;-w t-family: Calibri&quot;&gt;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;2.&amp;nbsp;&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important;font-family: 宋体;font-size: 14px;letter-spacing: 0.034em&quot;&gt;利用共享架构和训练过程进行分割任务：&lt;/span&gt;&lt;span style=&quot;-w t-family: Calibri&quot;&gt;Mask DINO&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important;font-family: 宋体;font-size: 14px;letter-spacing: 0.034em&quot;&gt;重用了&lt;/span&gt;&lt;span style=&quot;-w t-family: Calibri&quot;&gt;DINO&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important;font-family: 宋体;font-size: 14px;letter-spacing: 0.034em&quot;&gt;中的内容查询嵌入，通过点积操作进行掩码分类，并提出了三项关键改进（统一增强的查询选择、掩码的统一去噪训练、混合双边匹配）来提高分割性能。&lt;/span&gt;&lt;/p&gt;&lt;p style=&quot;-w t, &amp;#39;Helvetica Neue&amp;#39;, &amp;#39;Hiragino Sans GB&amp;#39;, &amp;#39;Microsoft YaHei UI&amp;#39;, &amp;#39;Microsoft YaHei&amp;#39;, Arial, sans-serif;font-size: 17px;letter-spacing: 0.544px;text-align: justify;text-wrap: wrap;background-color: rgb(255, 255, 255);box-sizing: border-box !important;overflow-wrap: break-word !important&quot;&gt;&lt;span style=&quot;-w t-family: Calibri&quot;&gt;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;3.&amp;nbsp;&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important;font-family: 宋体;font-size: 14px;letter-spacing: 0.034em&quot;&gt;证明检测和分割任务可以通过共享架构互相帮助：实验表明，特别是在复杂背景的分割中，检测任务可以显著提升分割任务的效果。&lt;/span&gt;&lt;/p&gt;&lt;p style=&quot;-w t, &amp;#39;Helvetica Neue&amp;#39;, &amp;#39;Hiragino Sans GB&amp;#39;, &amp;#39;Microsoft YaHei UI&amp;#39;, &amp;#39;Microsoft YaHei&amp;#39;, Arial, sans-serif;font-size: 17px;letter-spacing: 0.544px;text-align: justify;text-wrap: wrap;background-color: rgb(255, 255, 255);box-sizing: border-box !important;overflow-wrap: break-word !important&quot;&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important;font-family: 宋体;font-size: 14px&quot;&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important;font-family: Calibri&quot;&gt;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;4.&amp;nbsp;&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important&quot;&gt;引入掩码增强的锚框初始化&lt;/span&gt;&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important;font-family: 宋体;font-size: 14px&quot;&gt;方法&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important;font-family: 宋体;font-size: 14px&quot;&gt;：提出了一种新的查询选择方法，利用编码器中的密集特征更好地初始化锚框，提高了检测的精度和分割任务的协同效应。&lt;/span&gt;&lt;/p&gt;&lt;p style=&quot;-w t, &amp;#39;Helvetica Neue&amp;#39;, &amp;#39;Hiragino Sans GB&amp;#39;, &amp;#39;Microsoft YaHei UI&amp;#39;, &amp;#39;Microsoft YaHei&amp;#39;, Arial, sans-serif;font-size: 17px;letter-spacing: 0.544px;text-align: justify;text-wrap: wrap;background-color: rgb(255, 255, 255);box-sizing: border-box !important;overflow-wrap: break-word !important&quot;&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important;font-family: 宋体;font-size: 14px&quot;&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important;font-family: Calibri&quot;&gt;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;5.&amp;nbsp;&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important&quot;&gt;提出掩码的统一去噪训练：扩展了去噪训练方法到分割任务，加快了训练收敛速度并提升了分割性能。&lt;/span&gt;&lt;/span&gt;&lt;/p&gt;&lt;p style=&quot;-w t, &amp;#39;Helvetica Neue&amp;#39;, &amp;#39;Hiragino Sans GB&amp;#39;, &amp;#39;Microsoft YaHei UI&amp;#39;, &amp;#39;Microsoft YaHei&amp;#39;, Arial, sans-serif;font-size: 17px;letter-spacing: 0.544px;text-align: justify;text-wrap: wrap;background-color: rgb(255, 255, 255);box-sizing: border-box !important;overflow-wrap: break-word !important&quot;&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important;font-family: 宋体;font-size: 14px&quot;&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important;font-family: Calibri&quot;&gt;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;6.&amp;nbsp;&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important&quot;&gt;引入混合双边匹配方法：增加了掩码预测损失，优化了匹配结果的准确性和一致性，进一步提升了模型的表现。&lt;/span&gt;&lt;/span&gt;&lt;/p&gt;&lt;p style=&quot;-w t, &amp;#39;Helvetica Neue&amp;#39;, &amp;#39;Hiragino Sans GB&amp;#39;, &amp;#39;Microsoft YaHei UI&amp;#39;, &amp;#39;Microsoft YaHei&amp;#39;, Arial, sans-serif;font-size: 17px;letter-spacing: 0.544px;text-align: justify;text-wrap: wrap;background-color: rgb(255, 255, 255);box-sizing: border-box !important;overflow-wrap: break-word !important&quot;&gt;&lt;strong style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important&quot;&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important;font-family: 宋体;font-size: 14px&quot;&gt;三、&lt;/span&gt;&lt;/strong&gt;&lt;strong style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important&quot;&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important;font-family: 宋体;font-size: 14px&quot;&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important;font-family: Calibri&quot;&gt;Mask DINO&lt;/span&gt;&lt;/span&gt;&lt;/strong&gt;&lt;strong style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important&quot;&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important;font-family: 宋体;font-size: 14px&quot;&gt;实验&lt;/span&gt;&lt;/strong&gt;&lt;strong style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important&quot;&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important;font-family: 宋体;font-size: 14px&quot;&gt;结果&lt;/span&gt;&lt;/strong&gt;&lt;strong style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important&quot;&gt;&lt;/strong&gt;&lt;/p&gt;&lt;p style=&quot;-w t, &amp;#39;Helvetica Neue&amp;#39;, &amp;#39;Hiragino Sans GB&amp;#39;, &amp;#39;Microsoft YaHei UI&amp;#39;, &amp;#39;Microsoft YaHei&amp;#39;, Arial, sans-serif;font-size: 17px;letter-spacing: 0.544px;text-align: justify;text-wrap: wrap;background-color: rgb(255, 255, 255);box-sizing: border-box !important;overflow-wrap: break-word !important&quot;&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important;font-family: 宋体;font-size: 14px&quot;&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important;font-family: Calibri&quot;&gt;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;3.1&amp;nbsp;&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important&quot;&gt;实例分割和目标检测：&lt;/span&gt;&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important;font-family: 宋体;font-size: 14px;letter-spacing: 0.034em&quot;&gt;在&lt;/span&gt;&lt;span style=&quot;-w t-family: Calibri&quot;&gt;COCO&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important;font-family: 宋体;font-size: 14px;letter-spacing: 0.034em&quot;&gt;数据集上，&lt;/span&gt;&lt;span style=&quot;-w t-family: Calibri&quot;&gt;Mask DINO&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important;font-family: 宋体;font-size: 14px;letter-spacing: 0.034em&quot;&gt;相较于现有的实例分割和目标检测模型（如&lt;/span&gt;&lt;span style=&quot;-w t-family: Calibri&quot;&gt;Mask2Former&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important;font-family: 宋体;font-size: 14px;letter-spacing: 0.034em&quot;&gt;和&lt;/span&gt;&lt;span style=&quot;-w t-family: Calibri&quot;&gt;DINO&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important;font-family: 宋体;font-size: 14px;letter-spacing: 0.034em&quot;&gt;）取得了显著更高的性能。在&lt;/span&gt;&lt;span style=&quot;-w t-family: Calibri&quot;&gt;ResNet-50&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important;font-family: 宋体;font-size: 14px;letter-spacing: 0.034em&quot;&gt;主干网络上，&lt;/span&gt;&lt;span style=&quot;-w t-family: Calibri&quot;&gt;Mask DINO&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important;font-family: 宋体;font-size: 14px;letter-spacing: 0.034em&quot;&gt;在实例分割任务中实现了&lt;/span&gt;&lt;span style=&quot;-w t-family: Calibri&quot;&gt;+2.7 AP&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important;font-family: 宋体;font-size: 14px;letter-spacing: 0.034em&quot;&gt;的提升，并在&lt;/span&gt;&lt;span style=&quot;-w t-family: Calibri&quot;&gt;50&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important;font-family: 宋体;font-size: 14px;letter-spacing: 0.034em&quot;&gt;轮训练中达到了最佳的检测性能。使用&lt;/span&gt;&lt;span style=&quot;-w t-family: Calibri&quot;&gt;SwinL&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important;font-family: 宋体;font-size: 14px;letter-spacing: 0.034em&quot;&gt;主干网络时，&lt;/span&gt;&lt;span style=&quot;-w t-family: Calibri&quot;&gt;Mask DINO&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important;font-family: 宋体;font-size: 14px;letter-spacing: 0.034em&quot;&gt;也在所有检测和分割任务上达到了最佳结果。&lt;/span&gt;&lt;/p&gt;&lt;p style=&quot;-w t, &amp;#39;Helvetica Neue&amp;#39;, &amp;#39;Hiragino Sans GB&amp;#39;, &amp;#39;Microsoft YaHei UI&amp;#39;, &amp;#39;Microsoft YaHei&amp;#39;, Arial, sans-serif;font-size: 17px;letter-spacing: 0.544px;text-wrap: wrap;background-color: rgb(255, 255, 255);text-align: center;box-sizing: border-box !important;overflow-wrap: break-word !important&quot;&gt;&lt;img class=&quot;ue-image&quot; src=&quot;http://www.trizmed.cn/zb_users/upload/2024/08/202408311725090990454139.png&quot; title=&quot;图片2.png&quot; alt=&quot;图片2.png&quot;/&gt;&lt;/p&gt;&lt;p style=&quot;-w t, &amp;#39;Helvetica Neue&amp;#39;, &amp;#39;Hiragino Sans GB&amp;#39;, &amp;#39;Microsoft YaHei UI&amp;#39;, &amp;#39;Microsoft YaHei&amp;#39;, Arial, sans-serif;font-size: 17px;letter-spacing: 0.544px;text-wrap: wrap;background-color: rgb(255, 255, 255);text-align: center;box-sizing: border-box !important;overflow-wrap: break-word !important&quot;&gt;&lt;span style=&quot;-w t-family: 宋体&quot;&gt;图&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;font-family: Calibri;box-sizing: border-box !important;overflow-wrap: break-word !important&quot;&gt;2 Mask DNIO&amp;nbsp;&lt;/span&gt;在&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;font-family: Calibri;box-sizing: border-box !important;overflow-wrap: break-word !important&quot;&gt;COCO&lt;/span&gt;数据集上实例分割和目标检测任务表现&lt;/span&gt;&lt;/p&gt;&lt;p style=&quot;-w t, &amp;#39;Helvetica Neue&amp;#39;, &amp;#39;Hiragino Sans GB&amp;#39;, &amp;#39;Microsoft YaHei UI&amp;#39;, &amp;#39;Microsoft YaHei&amp;#39;, Arial, sans-serif;font-size: 17px;letter-spacing: 0.544px;text-align: justify;text-wrap: wrap;background-color: rgb(255, 255, 255);box-sizing: border-box !important;overflow-wrap: break-word !important&quot;&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important;font-family: 宋体;font-size: 14px&quot;&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important;font-family: Calibri&quot;&gt;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;3.2&amp;nbsp;&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important&quot;&gt;全景分割：在&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important;font-family: Calibri&quot;&gt;COCO&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important&quot;&gt;数据集的全景分割任务中，&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important;font-family: Calibri&quot;&gt;Mask DINO&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important&quot;&gt;超越了所有现有模型，并在&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important;font-family: Calibri&quot;&gt;12&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important&quot;&gt;和&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important;font-family: Calibri&quot;&gt;50&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important&quot;&gt;轮训练中分别提高了&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important;font-family: Calibri&quot;&gt;1.0 PQ&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important&quot;&gt;和&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important;font-family: Calibri&quot;&gt;1.1 PQ&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important&quot;&gt;。&lt;/span&gt;&lt;/span&gt;&lt;/p&gt;&lt;p style=&quot;-w t, &amp;#39;Helvetica Neue&amp;#39;, &amp;#39;Hiragino Sans GB&amp;#39;, &amp;#39;Microsoft YaHei UI&amp;#39;, &amp;#39;Microsoft YaHei&amp;#39;, Arial, sans-serif;font-size: 17px;letter-spacing: 0.544px;text-wrap: wrap;background-color: rgb(255, 255, 255);text-align: center;box-sizing: border-box !important;overflow-wrap: break-word !important&quot;&gt;&lt;img class=&quot;ue-image&quot; src=&quot;http://www.trizmed.cn/zb_users/upload/2024/08/202408311725091003468082.png&quot; title=&quot;图片3.png&quot; alt=&quot;图片3.png&quot;/&gt;&lt;/p&gt;&lt;p style=&quot;-w t, &amp;#39;Helvetica Neue&amp;#39;, &amp;#39;Hiragino Sans GB&amp;#39;, &amp;#39;Microsoft YaHei UI&amp;#39;, &amp;#39;Microsoft YaHei&amp;#39;, Arial, sans-serif;font-size: 17px;letter-spacing: 0.544px;text-wrap: wrap;background-color: rgb(255, 255, 255);text-align: center;box-sizing: border-box !important;overflow-wrap: break-word !important&quot;&gt;&lt;span style=&quot;-w t-family: 宋体&quot;&gt;图&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;font-family: Calibri;box-sizing: border-box !important;overflow-wrap: break-word !important&quot;&gt;3 Mask DNIO&amp;nbsp;&lt;/span&gt;在&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;font-family: Calibri;box-sizing: border-box !important;overflow-wrap: break-word !important&quot;&gt;COCO&lt;/span&gt;数据集上全景分割任务表现&lt;/span&gt;&lt;/p&gt;&lt;p style=&quot;-w t, &amp;#39;Helvetica Neue&amp;#39;, &amp;#39;Hiragino Sans GB&amp;#39;, &amp;#39;Microsoft YaHei UI&amp;#39;, &amp;#39;Microsoft YaHei&amp;#39;, Arial, sans-serif;font-size: 17px;letter-spacing: 0.544px;text-align: justify;text-wrap: wrap;background-color: rgb(255, 255, 255);box-sizing: border-box !important;overflow-wrap: break-word !important&quot;&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important;font-family: 宋体;font-size: 14px&quot;&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important;font-family: Calibri&quot;&gt;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;3.3&amp;nbsp;&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important&quot;&gt;语义分割：在&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important;font-family: Calibri&quot;&gt;ADE20K&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important&quot;&gt;和&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important;font-family: Calibri&quot;&gt;Cityscapes&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important&quot;&gt;数据集上，&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important;font-family: Calibri&quot;&gt;Mask DINO&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important&quot;&gt;在语义分割任务中表现出色，比现有的&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important;font-family: Calibri&quot;&gt;Mask2Former&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important&quot;&gt;模型分别高出&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important;font-family: Calibri&quot;&gt;1.6&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important&quot;&gt;和&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important;font-family: Calibri&quot;&gt;0.6 mIoU&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important&quot;&gt;。&lt;/span&gt;&lt;/span&gt;&lt;/p&gt;&lt;p style=&quot;-w t, &amp;#39;Helvetica Neue&amp;#39;, &amp;#39;Hiragino Sans GB&amp;#39;, &amp;#39;Microsoft YaHei UI&amp;#39;, &amp;#39;Microsoft YaHei&amp;#39;, Arial, sans-serif;font-size: 17px;letter-spacing: 0.544px;text-wrap: wrap;background-color: rgb(255, 255, 255);text-align: center;box-sizing: border-box !important;overflow-wrap: break-word !important&quot;&gt;&lt;img class=&quot;ue-image&quot; src=&quot;http://www.trizmed.cn/zb_users/upload/2024/08/202408311725091021597295.png&quot; title=&quot;图片4.png&quot; alt=&quot;图片4.png&quot;/&gt;&lt;/p&gt;&lt;p style=&quot;-w t, &amp;#39;Helvetica Neue&amp;#39;, &amp;#39;Hiragino Sans GB&amp;#39;, &amp;#39;Microsoft YaHei UI&amp;#39;, &amp;#39;Microsoft YaHei&amp;#39;, Arial, sans-serif;font-size: 17px;letter-spacing: 0.544px;text-wrap: wrap;background-color: rgb(255, 255, 255);text-align: center;box-sizing: border-box !important;overflow-wrap: break-word !important&quot;&gt;&lt;span style=&quot;-w t-family: 宋体&quot;&gt;图&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;font-family: Calibri;box-sizing: border-box !important;overflow-wrap: break-word !important&quot;&gt;4 Mask DNIO&amp;nbsp;&lt;/span&gt;在&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;font-family: Calibri;box-sizing: border-box !important;overflow-wrap: break-word !important&quot;&gt;COCO&lt;/span&gt;数据集上语义分割任务表现&lt;/span&gt;&lt;/p&gt;&lt;p style=&quot;-w t, &amp;#39;Helvetica Neue&amp;#39;, &amp;#39;Hiragino Sans GB&amp;#39;, &amp;#39;Microsoft YaHei UI&amp;#39;, &amp;#39;Microsoft YaHei&amp;#39;, Arial, sans-serif;font-size: 17px;letter-spacing: 0.544px;text-align: justify;text-wrap: wrap;background-color: rgb(255, 255, 255);box-sizing: border-box !important;overflow-wrap: break-word !important&quot;&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important;font-family: 宋体;font-size: 14px&quot;&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important;font-family: Calibri&quot;&gt;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;3.4&amp;nbsp;&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important&quot;&gt;与&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important;font-family: Calibri&quot;&gt;SOTA&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important&quot;&gt;模型的比较：&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important;font-family: Calibri&quot;&gt;Mask DINO&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important&quot;&gt;在不使用额外数据的情况下，超过了所有现有模型的性能，并在使用大规模检测数据集（如&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important;font-family: Calibri&quot;&gt;Objects365&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important&quot;&gt;）进行预训练后，进一步提升了所有分割任务的表现。&lt;/span&gt;&lt;/span&gt;&lt;/p&gt;&lt;p style=&quot;-w t, &amp;#39;Helvetica Neue&amp;#39;, &amp;#39;Hiragino Sans GB&amp;#39;, &amp;#39;Microsoft YaHei UI&amp;#39;, &amp;#39;Microsoft YaHei&amp;#39;, Arial, sans-serif;font-size: 17px;letter-spacing: 0.544px;text-wrap: wrap;background-color: rgb(255, 255, 255);text-align: center;box-sizing: border-box !important;overflow-wrap: break-word !important&quot;&gt;&lt;img class=&quot;ue-image&quot; src=&quot;http://www.trizmed.cn/zb_users/upload/2024/08/202408311725091031448814.png&quot; title=&quot;图片5.png&quot; alt=&quot;图片5.png&quot;/&gt;&lt;/p&gt;&lt;p style=&quot;-w t, &amp;#39;Helvetica Neue&amp;#39;, &amp;#39;Hiragino Sans GB&amp;#39;, &amp;#39;Microsoft YaHei UI&amp;#39;, &amp;#39;Microsoft YaHei&amp;#39;, Arial, sans-serif;font-size: 17px;letter-spacing: 0.544px;text-wrap: wrap;background-color: rgb(255, 255, 255);text-align: center;box-sizing: border-box !important;overflow-wrap: break-word !important&quot;&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important;font-size: 12px&quot;&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;font-family: 宋体;letter-spacing: 0.034em;box-sizing: border-box !important;overflow-wrap: break-word !important&quot;&gt;图&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;letter-spacing: 0.034em;font-family: Calibri;box-sizing: border-box !important;overflow-wrap: break-word !important&quot;&gt;5 Mask DNIO&amp;nbsp;&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;font-family: 宋体;letter-spacing: 0.034em;box-sizing: border-box !important;overflow-wrap: break-word !important&quot;&gt;与当前&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;letter-spacing: 0.034em;font-family: Calibri;box-sizing: border-box !important;overflow-wrap: break-word !important&quot;&gt;SOTA&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;font-family: 宋体;letter-spacing: 0.034em;box-sizing: border-box !important;overflow-wrap: break-word !important&quot;&gt;模型比较结果&lt;/span&gt;&lt;/span&gt;&lt;/p&gt;&lt;p style=&quot;-w t, &amp;#39;Helvetica Neue&amp;#39;, &amp;#39;Hiragino Sans GB&amp;#39;, &amp;#39;Microsoft YaHei UI&amp;#39;, &amp;#39;Microsoft YaHei&amp;#39;, Arial, sans-serif;font-size: 17px;letter-spacing: 0.544px;text-align: justify;text-wrap: wrap;background-color: rgb(255, 255, 255);box-sizing: border-box !important;overflow-wrap: break-word !important&quot;&gt;&lt;strong style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important&quot;&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important;font-family: 宋体;font-size: 14px&quot;&gt;四、总结&lt;/span&gt;&lt;/strong&gt;&lt;strong style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important&quot;&gt;&lt;/strong&gt;&lt;/p&gt;&lt;p style=&quot;-w t, &amp;#39;Helvetica Neue&amp;#39;, &amp;#39;Hiragino Sans GB&amp;#39;, &amp;#39;Microsoft YaHei UI&amp;#39;, &amp;#39;Microsoft YaHei&amp;#39;, Arial, sans-serif;font-size: 17px;letter-spacing: 0.544px;text-align: justify;text-wrap: wrap;background-color: rgb(255, 255, 255);box-sizing: border-box !important;overflow-wrap: break-word !important&quot;&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important;font-family: 宋体;font-size: 14px&quot;&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important;font-family: Calibri&quot;&gt;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;Mask DINO&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important&quot;&gt;框架成功地统一了目标检测和图像分割任务，通过引入掩码预测分支和共享架构，显著提升了各种分割任务的性能。&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important;font-family: Calibri&quot;&gt;Mask DINO&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important&quot;&gt;不仅在多个数据集上取得了当前最优的结果，还展示了检测和分割任务在共享架构中的互助潜力。&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important;font-family: Calibri&quot;&gt;Mask DINO&lt;/span&gt;&lt;span style=&quot;-webkit-tap-highlight-color: transparent;padding: 0px;outline: 0px;max-width: 100%;box-sizing: border-box !important;overflow-wrap: break-word !important&quot;&gt;的成功为更多视觉任务的统一框架设计和任务协作提供了新的研究方向，如集成跟踪、姿态估计等任务，从而推动多任务联合学习的发展。&lt;/span&gt;&lt;/span&gt;&lt;/p&gt;&lt;p&gt;&lt;br/&gt;&lt;/p&gt;</description><pubDate>Sat, 31 Aug 2024 15:53:29 +0800</pubDate></item><item><title>第四期-智能机器人辅助诊疗方法、系统、智能机器人及存储介质</title><link>http://www.trizmed.cn/jscxd/11.html</link><description>&lt;h1 style=&quot;box-sizing: border-box; margin: 23px -30px 22px; padding: 10px 0px 10px 40px; outline: 0px; border-width: 0px 0px 0px 4px; border-top-style: initial; border-right-style: initial; border-bottom-style: initial; border-left-style: solid; border-top-color: initial; border-right-color: initial; border-bottom-color: initial; border-left-color: rgb(193, 30, 34); border-image: initial; vertical-align: baseline; font-size: 29px; background: rgb(245, 245, 245); color: rgb(68, 68, 68); text-wrap: wrap; break-after: avoid; text-align: justify; line-height: 70.4px; font-family: 等线;&quot;&gt;&lt;strong style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; color: rgb(102, 102, 102); overflow-wrap: break-word;&quot;&gt;1.&amp;nbsp;&lt;/strong&gt;&lt;strong style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; color: rgb(102, 102, 102); overflow-wrap: break-word;&quot;&gt;智能机器人辅助诊疗方法、系统、智能机器人及存储介质&lt;/strong&gt;&lt;/h1&gt;&lt;p style=&quot;box-sizing: border-box; margin-top: 0px; margin-bottom: 10px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 15px; line-height: 27px; color: rgb(102, 102, 102); text-align: justify; text-wrap: wrap; font-family: 等线; background: rgb(255, 255, 255);&quot;&gt;&lt;strong style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; overflow-wrap: break-word;&quot;&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px; font-family: 宋体;&quot;&gt;申请人&lt;/span&gt;&lt;/strong&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px; font-family: 宋体;&quot;&gt;：中国人民解放军总医院&lt;/span&gt;&lt;/p&gt;&lt;p style=&quot;box-sizing: border-box; margin-top: 0px; margin-bottom: 10px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 15px; line-height: 27px; color: rgb(102, 102, 102); text-align: justify; text-wrap: wrap; font-family: 等线; background: rgb(255, 255, 255);&quot;&gt;&lt;strong style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; overflow-wrap: break-word;&quot;&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px; font-family: 宋体;&quot;&gt;申请号&lt;/span&gt;&lt;/strong&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px; font-family: 宋体;&quot;&gt;：&lt;/span&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px; font-family: 宋体;&quot;&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-family: Arial;&quot;&gt;202311444378.6&lt;/span&gt;&lt;/span&gt;&lt;/p&gt;&lt;p style=&quot;box-sizing: border-box; margin-top: 0px; margin-bottom: 10px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 15px; line-height: 27px; color: rgb(102, 102, 102); text-align: justify; text-wrap: wrap; font-family: 等线; background: rgb(255, 255, 255);&quot;&gt;&lt;strong style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; overflow-wrap: break-word;&quot;&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px; font-family: 宋体;&quot;&gt;申请日&lt;/span&gt;&lt;/strong&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px; font-family: 宋体;&quot;&gt;：&lt;/span&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px; font-family: 宋体;&quot;&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-family: Arial;&quot;&gt;2023.11.01&lt;/span&gt;&lt;/span&gt;&lt;/p&gt;&lt;h2 style=&quot;box-sizing: border-box; margin: 17px -30px; padding: 10px 0px 10px 40px; outline: 0px; border-width: 0px 0px 0px 4px; border-top-style: initial; border-right-style: initial; border-bottom-style: initial; border-left-style: solid; border-top-color: initial; border-right-color: initial; border-bottom-color: initial; border-left-color: rgb(70, 147, 233); border-image: initial; vertical-align: baseline; font-size: 21px; background: rgb(245, 245, 245); color: rgb(68, 68, 68); text-wrap: wrap; break-after: avoid; text-align: justify; line-height: 36.9067px; font-family: &amp;quot;等线 Light&amp;quot;;&quot;&gt;&lt;strong style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; color: rgb(102, 102, 102); overflow-wrap: break-word;&quot;&gt;1.1 发明问题&lt;/strong&gt;&lt;/h2&gt;&lt;h2 style=&quot;box-sizing: border-box; margin: 17px -30px; padding: 10px 0px 10px 40px; outline: 0px; border-width: 0px 0px 0px 4px; border-top-style: initial; border-right-style: initial; border-bottom-style: initial; border-left-style: solid; border-top-color: initial; border-right-color: initial; border-bottom-color: initial; border-left-color: rgb(70, 147, 233); border-image: initial; vertical-align: baseline; font-size: 21px; background: rgb(245, 245, 245); color: rgb(68, 68, 68); text-wrap: wrap; break-after: avoid; text-align: justify; line-height: 36.9067px; font-family: &amp;quot;等线 Light&amp;quot;;&quot;&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px; font-family: 宋体; font-weight: normal;&quot;&gt;实际应用中，医生亲自对患者进行问诊，并基于问诊结果确定患者进一步的诊疗方案。这种诊疗方式的效率不高，耗费了医生的大量时间，特别是在医疗资源紧张，住院病人多，医生稀缺的场景下，还容易出现诊疗失误。&lt;/span&gt;&lt;/h2&gt;&lt;h2 style=&quot;box-sizing: border-box; margin: 17px -30px; padding: 10px 0px 10px 40px; outline: 0px; border-width: 0px 0px 0px 4px; border-top-style: initial; border-right-style: initial; border-bottom-style: initial; border-left-style: solid; border-top-color: initial; border-right-color: initial; border-bottom-color: initial; border-left-color: rgb(70, 147, 233); border-image: initial; vertical-align: baseline; font-size: 21px; background: rgb(245, 245, 245); color: rgb(68, 68, 68); text-wrap: wrap; break-after: avoid; text-align: justify; line-height: 36.9067px; font-family: &amp;quot;等线 Light&amp;quot;;&quot;&gt;&lt;strong style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; color: rgb(102, 102, 102); overflow-wrap: break-word;&quot;&gt;1.2 本发明技术方案&lt;/strong&gt;&lt;/h2&gt;&lt;p style=&quot;box-sizing: border-box; margin-top: 0px; margin-right: 16px; margin-bottom: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 15px; line-height: 27px; color: rgb(102, 102, 102); text-align: center; text-wrap: wrap; font-family: 等线; background: rgb(255, 255, 255);&quot;&gt;&lt;img src=&quot;http://www.trizmed.cn/zb_users/upload/2024/08/202408291724922126737645.png&quot; alt=&quot;image.png&quot;/&gt;&lt;/p&gt;&lt;p style=&quot;box-sizing: border-box; margin-top: 0px; margin-bottom: 10px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 15px; line-height: 27px; color: rgb(102, 102, 102); text-align: justify; text-wrap: wrap; font-family: 等线; background: rgb(255, 255, 255);&quot;&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px; font-family: 宋体;&quot;&gt;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;发明方案&lt;/span&gt;&lt;/p&gt;&lt;p style=&quot;box-sizing: border-box; margin-top: 0px; margin-bottom: 10px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 15px; line-height: 27px; color: rgb(102, 102, 102); text-align: justify; text-wrap: wrap; font-family: 等线; background: rgb(255, 255, 255);&quot;&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px; font-family: 宋体;&quot;&gt;&amp;nbsp;&lt;/span&gt;&lt;/p&gt;&lt;h2 style=&quot;box-sizing: border-box; margin: 17px -30px; padding: 10px 0px 10px 40px; outline: 0px; border-width: 0px 0px 0px 4px; border-top-style: initial; border-right-style: initial; border-bottom-style: initial; border-left-style: solid; border-top-color: initial; border-right-color: initial; border-bottom-color: initial; border-left-color: rgb(70, 147, 233); border-image: initial; vertical-align: baseline; font-size: 21px; background: rgb(245, 245, 245); color: rgb(68, 68, 68); text-wrap: wrap; break-after: avoid; text-align: justify; line-height: 36.9067px; font-family: &amp;quot;等线 Light&amp;quot;;&quot;&gt;&lt;strong style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; color: rgb(102, 102, 102); overflow-wrap: break-word;&quot;&gt;1.3 创新点&lt;/strong&gt;&lt;/h2&gt;&lt;p style=&quot;box-sizing: border-box; margin-top: 0px; margin-bottom: 10px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 15px; line-height: 27px; color: rgb(102, 102, 102); text-align: justify; text-wrap: wrap; font-family: 等线; background: rgb(255, 255, 255);&quot;&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px; font-family: 宋体;&quot;&gt;本发明&lt;/span&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px; font-family: 宋体;&quot;&gt;首先识别患者身份信息，然后去云端获取该患者历史病例信息；然后利用医疗大语言模型根据患者病例信息输出针对该患者的问诊问题；利用这些问诊问题，大模型自动同患者进行多轮对话，对话过程中记录问诊结果并存到患者的电子病例中。&lt;/span&gt;&lt;/p&gt;&lt;p style=&quot;box-sizing: border-box; margin-top: 0px; margin-right: 16px; margin-bottom: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 15px; line-height: 27px; color: rgb(102, 102, 102); text-align: justify; text-wrap: wrap; font-family: 等线; background: rgb(255, 255, 255);&quot;&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px; font-family: 宋体;&quot;&gt;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&lt;/span&gt;&lt;/p&gt;&lt;h2 style=&quot;box-sizing: border-box; margin: 17px -30px; padding: 10px 0px 10px 40px; outline: 0px; border-width: 0px 0px 0px 4px; border-top-style: initial; border-right-style: initial; border-bottom-style: initial; border-left-style: solid; border-top-color: initial; border-right-color: initial; border-bottom-color: initial; border-left-color: rgb(70, 147, 233); border-image: initial; vertical-align: baseline; font-size: 21px; background: rgb(245, 245, 245); color: rgb(68, 68, 68); text-wrap: wrap; break-after: avoid; text-align: justify; line-height: 36.9067px; font-family: &amp;quot;等线 Light&amp;quot;;&quot;&gt;&lt;strong style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; color: rgb(102, 102, 102); overflow-wrap: break-word;&quot;&gt;1.4 本发明用到的工具和方法：&lt;/strong&gt;&lt;/h2&gt;&lt;h3 style=&quot;box-sizing: border-box; margin: 17px -30px; padding: 7px 0px 6px 40px; outline: 0px; border-width: 0px 0px 0px 4px; border-top-style: initial; border-right-style: initial; border-bottom-style: initial; border-left-style: solid; border-top-color: initial; border-right-color: initial; border-bottom-color: initial; border-left-color: rgb(255, 174, 73); border-image: initial; vertical-align: baseline; font-size: 21px; color: rgb(85, 85, 85); background: rgb(245, 245, 245); text-wrap: wrap; break-after: avoid; text-align: justify; line-height: 36.9067px; font-family: 等线;&quot;&gt;&lt;strong style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; color: rgb(102, 102, 102); overflow-wrap: break-word;&quot;&gt;1.4.1&amp;nbsp;&lt;/strong&gt;&lt;strong style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; color: rgb(102, 102, 102); overflow-wrap: break-word;&quot;&gt;大语言模型&lt;/strong&gt;&lt;strong style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; color: rgb(102, 102, 102); overflow-wrap: break-word;&quot;&gt;:&lt;/strong&gt;&lt;/h3&gt;&lt;p style=&quot;box-sizing: border-box; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 15px; line-height: 27px; color: rgb(102, 102, 102); text-align: justify; text-wrap: wrap; background-color: rgb(255, 255, 255); font-family: 等线;&quot;&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px; font-family: 宋体; background: #FFFF00;&quot;&gt;专利中涉及到的大语言模型简单理解就是一个支持多轮对话的百科全书，专利中涉及到的医疗大语言模型简单理解就是一个支持多轮对话的医疗百科全书&lt;/span&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px; font-family: Arial; background: #FFFF00;&quot;&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-family: 宋体;&quot;&gt;。&lt;/span&gt;&lt;/span&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px; font-family: 宋体; background: #FFFF00;&quot;&gt;大语言模型的详细介绍如下：&lt;/span&gt;&lt;/p&gt;&lt;p style=&quot;box-sizing: border-box; margin-top: 0px; margin-bottom: 21px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 15px; line-height: 27px; color: rgb(102, 102, 102); text-align: justify; text-wrap: wrap; background-color: rgb(255, 255, 255); font-family: 等线;&quot;&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px;&quot;&gt;大语言模型是指一类能够理解和生成自然语言文本的计算机程序或系统。这些模型通常基于深度学习技术，尤其是循环神经网络（RNN）或变换器（Transformer）等架构。这些模型在大规模文本数据上进行训练，以学习语言的结构、语法、语义和上下文关系，从而能够执行多种自然语言处理任务，如文本生成、机器翻译、语言理解等。例如，像ChatGPT这样的大语言模型能够与人类进行自然对话，并提供有用的信息和建议。&lt;/span&gt;&lt;/p&gt;&lt;h3 style=&quot;box-sizing: border-box; margin: 17px -30px; padding: 7px 0px 6px 40px; outline: 0px; border-width: 0px 0px 0px 4px; border-top-style: initial; border-right-style: initial; border-bottom-style: initial; border-left-style: solid; border-top-color: initial; border-right-color: initial; border-bottom-color: initial; border-left-color: rgb(255, 174, 73); border-image: initial; vertical-align: baseline; font-size: 21px; color: rgb(85, 85, 85); background: rgb(245, 245, 245); text-wrap: wrap; break-after: avoid; text-align: justify; line-height: 36.9067px; font-family: 等线;&quot;&gt;&lt;strong style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; color: rgb(102, 102, 102); overflow-wrap: break-word;&quot;&gt;1.4.2&amp;nbsp;&lt;/strong&gt;&lt;strong style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; color: rgb(102, 102, 102); overflow-wrap: break-word;&quot;&gt;知识图谱&lt;/strong&gt;&lt;/h3&gt;&lt;p style=&quot;box-sizing: border-box; margin-top: 0px; margin-bottom: 21px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 15px; line-height: 27px; color: rgb(102, 102, 102); text-align: justify; text-wrap: wrap; background-color: rgb(255, 255, 255); font-family: 等线;&quot;&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px; font-family: 宋体; background: #FFFF00;&quot;&gt;专利中对知识图谱的介绍基本没有，这一概念在这个专利中没有意义可忽略。知识图谱这个概念的具体介绍如下：&lt;/span&gt;&lt;/p&gt;&lt;p style=&quot;box-sizing: border-box; margin-top: 0px; margin-bottom: 21px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 15px; line-height: 27px; color: rgb(102, 102, 102); text-align: justify; text-wrap: wrap; background-color: rgb(255, 255, 255); font-family: 等线;&quot;&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px;&quot;&gt;知识图谱是一种结构化的知识表示形式，用于描述现实世界中的实体、概念和它们之间的关系。它是一个语义网络，由节点（实体或概念）和边（关系）组成，用于表示知识之间的关联。知识图谱可以用于组织、存储和查询大规模的知识数据，并支持各种知识推理和分析任务。&lt;/span&gt;&lt;/p&gt;&lt;p style=&quot;box-sizing: border-box; margin-top: 0px; margin-bottom: 21px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 15px; line-height: 27px; color: rgb(102, 102, 102); text-align: justify; text-wrap: wrap; background-color: rgb(255, 255, 255); font-family: 等线;&quot;&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px;&quot;&gt;知识图谱的应用非常广泛，其中最著名的例子可能就是Google知识图谱，它用于改善搜索结果的准确性和相关性。另外，知识图谱还被用于推荐系统、智能个人助理、自然语言理解、语义搜索等领域。知识图谱的建立通常需要从多个来源收集、整合和验证数据，以构建一个全面和准确的知识库&lt;/span&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px;&quot;&gt;。&lt;/span&gt;&lt;/p&gt;&lt;h3 style=&quot;box-sizing: border-box; margin: 17px -30px; padding: 7px 0px 6px 40px; outline: 0px; border-width: 0px 0px 0px 4px; border-top-style: initial; border-right-style: initial; border-bottom-style: initial; border-left-style: solid; border-top-color: initial; border-right-color: initial; border-bottom-color: initial; border-left-color: rgb(255, 174, 73); border-image: initial; vertical-align: baseline; font-size: 21px; color: rgb(85, 85, 85); background: rgb(245, 245, 245); text-wrap: wrap; break-after: avoid; text-align: justify; line-height: 36.9067px; font-family: 等线;&quot;&gt;&lt;strong style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; color: rgb(102, 102, 102); overflow-wrap: break-word;&quot;&gt;1.4.3&amp;nbsp;&lt;/strong&gt;&lt;strong style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; color: rgb(102, 102, 102); overflow-wrap: break-word;&quot;&gt;云端&lt;/strong&gt;&lt;/h3&gt;&lt;p style=&quot;box-sizing: border-box; margin-top: 0px; margin-bottom: 21px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 15px; line-height: 27px; color: rgb(102, 102, 102); text-align: justify; text-wrap: wrap; background-color: rgb(255, 255, 255); font-family: 等线;&quot;&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px;&quot;&gt;指提供云服务的平台，&amp;nbsp;可以是硬件也可以是软件。当云端为硬件时，&amp;nbsp;该云端为单个服务器或多个服务器组成的分布式服务器集群。当云端为软件时，&amp;nbsp;可以为多个软件模块或单个软件模块等&lt;/span&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px;&quot;&gt;。&lt;/span&gt;&lt;/p&gt;&lt;h2 style=&quot;box-sizing: border-box; margin: 17px -30px; padding: 10px 0px 10px 40px; outline: 0px; border-width: 0px 0px 0px 4px; border-top-style: initial; border-right-style: initial; border-bottom-style: initial; border-left-style: solid; border-top-color: initial; border-right-color: initial; border-bottom-color: initial; border-left-color: rgb(70, 147, 233); border-image: initial; vertical-align: baseline; font-size: 21px; background: rgb(245, 245, 245); color: rgb(68, 68, 68); text-wrap: wrap; break-after: avoid; text-align: justify; line-height: 36.9067px; font-family: &amp;quot;等线 Light&amp;quot;;&quot;&gt;&lt;strong style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; color: rgb(102, 102, 102); overflow-wrap: break-word;&quot;&gt;1.5 结合创新工具的思考&lt;/strong&gt;&lt;/h2&gt;&lt;p style=&quot;box-sizing: border-box; margin-top: 0px; margin-bottom: 21px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 15px; line-height: 27px; color: rgb(102, 102, 102); text-align: justify; text-wrap: wrap; background-color: rgb(255, 255, 255); font-family: 等线;&quot;&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px; font-family: 宋体; background: #FFFF00;&quot;&gt;这个专利根据患者的历史就诊信息来进行问诊，存在以下缺点：&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-family: Arial;&quot;&gt;1&lt;/span&gt;）系统在设计时没有充分考虑到快速响应和适应用户或环境变化的能力，它可能在紧急医疗情况下反应不够迅速。例如，对于突发的医疗状况，系统需要能够即时识别并调整问诊策略，若处理不当，可能延误诊断；&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-family: Arial;&quot;&gt;2&lt;/span&gt;）系统的设计可能未能充分利用患者的实时健康数据来动态调整诊疗过程。如果系统主要依赖历史病例数据而不是实时数据，它可能无法准确反映患者当前的健康状况。&lt;/span&gt;&lt;/p&gt;&lt;p style=&quot;box-sizing: border-box; margin-top: 0px; margin-bottom: 21px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 15px; line-height: 27px; color: rgb(102, 102, 102); text-align: justify; text-wrap: wrap; background-color: rgb(255, 255, 255); font-family: 等线;&quot;&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px; font-family: 宋体; background: #FFFF00;&quot;&gt;针对上述不足，我们可以利用&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-family: Arial;&quot;&gt;40&lt;/span&gt;发明原理中的第&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-family: Arial;&quot;&gt;15&lt;/span&gt;条——动态特性对其进行创新，具体来说：系统可以设计为根据患者的实时健康数据动态调整问诊问题。例如，如果患者的血压突然升高，系统可以自动调整问诊重点，更加关注可能导致血压变化的因素。&lt;/span&gt;&lt;/p&gt;&lt;h1 style=&quot;box-sizing: border-box; margin: 23px -30px 22px; padding: 10px 0px 10px 40px; outline: 0px; border-width: 0px 0px 0px 4px; border-top-style: initial; border-right-style: initial; border-bottom-style: initial; border-left-style: solid; border-top-color: initial; border-right-color: initial; border-bottom-color: initial; border-left-color: rgb(193, 30, 34); border-image: initial; vertical-align: baseline; font-size: 29px; background: rgb(245, 245, 245); color: rgb(68, 68, 68); text-wrap: wrap; break-after: avoid; text-align: justify; line-height: 70.4px; font-family: 等线;&quot;&gt;&lt;strong style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; color: rgb(102, 102, 102); overflow-wrap: break-word;&quot;&gt;2.&amp;nbsp;&lt;/strong&gt;&lt;strong style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; color: rgb(102, 102, 102); overflow-wrap: break-word;&quot;&gt;一种运动训练损伤智能检测识别系统及方法&lt;/strong&gt;&lt;/h1&gt;&lt;p style=&quot;box-sizing: border-box; margin-top: 0px; margin-bottom: 10px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 15px; line-height: 27px; color: rgb(102, 102, 102); text-align: justify; text-wrap: wrap; font-family: 等线; background: rgb(255, 255, 255);&quot;&gt;&lt;strong style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; overflow-wrap: break-word;&quot;&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px; font-family: 宋体;&quot;&gt;申请人&lt;/span&gt;&lt;/strong&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px; font-family: 宋体;&quot;&gt;：中国人民解放军总医院&lt;/span&gt;&lt;/p&gt;&lt;p style=&quot;box-sizing: border-box; margin-top: 0px; margin-bottom: 10px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 15px; line-height: 27px; color: rgb(102, 102, 102); text-align: justify; text-wrap: wrap; font-family: 等线; background: rgb(255, 255, 255);&quot;&gt;&lt;strong style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; overflow-wrap: break-word;&quot;&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px; font-family: 宋体;&quot;&gt;申请号&lt;/span&gt;&lt;/strong&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px; font-family: 宋体;&quot;&gt;：&lt;/span&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px; font-family: 宋体;&quot;&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-family: Arial;&quot;&gt;202311747299.2&lt;/span&gt;&lt;/span&gt;&lt;/p&gt;&lt;p style=&quot;box-sizing: border-box; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 15px; line-height: 27px; color: rgb(102, 102, 102); text-align: justify; text-wrap: wrap; background-color: rgb(255, 255, 255); font-family: 等线;&quot;&gt;&lt;strong style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; overflow-wrap: break-word;&quot;&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px; font-family: 宋体;&quot;&gt;申请日&lt;/span&gt;&lt;/strong&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px; font-family: 宋体;&quot;&gt;：&lt;/span&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px; font-family: 宋体;&quot;&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-family: Arial;&quot;&gt;2023.12.19&lt;/span&gt;&lt;/span&gt;&lt;/p&gt;&lt;h2 style=&quot;box-sizing: border-box; margin: 17px -30px; padding: 10px 0px 10px 40px; outline: 0px; border-width: 0px 0px 0px 4px; border-top-style: initial; border-right-style: initial; border-bottom-style: initial; border-left-style: solid; border-top-color: initial; border-right-color: initial; border-bottom-color: initial; border-left-color: rgb(70, 147, 233); border-image: initial; vertical-align: baseline; font-size: 21px; background: rgb(245, 245, 245); color: rgb(68, 68, 68); text-wrap: wrap; break-after: avoid; text-align: justify; line-height: 36.9067px; font-family: &amp;quot;等线 Light&amp;quot;;&quot;&gt;&lt;strong style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; color: rgb(102, 102, 102); overflow-wrap: break-word;&quot;&gt;2.1 发明问题&lt;/strong&gt;&lt;/h2&gt;&lt;p style=&quot;box-sizing: border-box; margin-top: 0px; margin-bottom: 10px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 15px; line-height: 27px; color: rgb(102, 102, 102); text-align: justify; text-wrap: wrap; font-family: 等线; background: rgb(255, 255, 255);&quot;&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px; font-family: 宋体;&quot;&gt;公开号为&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-family: Arial;&quot;&gt;CN106990835B&lt;/span&gt;的中国专利公开了一种运动训练评估方法及装置， 主要通过检测和分析受试者进行运动训练时的脑电信号变化从而对交互系统中的触觉反馈进行定量化评估。解决了现有评估手段无法提供直观、定量的运动训练效果评价指标的问题， 能方便地应用于融合视&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-family: Arial;&quot;&gt;/&lt;/span&gt;触觉反馈的三维虚拟训练系统中， 弥补了诸如问卷调查或行为测量等现有评估手段的不足， 上述专利虽然解决了损伤检测的问题， 但是在实际操作中还存在以下问题：&lt;/span&gt;&lt;/p&gt;&lt;p style=&quot;box-sizing: border-box; margin-top: 0px; margin-bottom: 10px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 15px; line-height: 27px; color: rgb(102, 102, 102); text-align: justify; text-wrap: wrap; font-family: 等线; background: rgb(255, 255, 255);&quot;&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px; font-family: 宋体;&quot;&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-family: Arial;&quot;&gt;1.&lt;/span&gt;当损伤检测完成后没有根据检测结果进行进一步的康复计划制定和训练计划&lt;/span&gt;&lt;/p&gt;&lt;p style=&quot;box-sizing: border-box; margin-top: 0px; margin-bottom: 10px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 15px; line-height: 27px; color: rgb(102, 102, 102); text-align: justify; text-wrap: wrap; font-family: 等线; background: rgb(255, 255, 255);&quot;&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px; font-family: 宋体;&quot;&gt;制定，&amp;nbsp;从而导致运动员后期训练和康复效果不佳。&lt;/span&gt;&lt;/p&gt;&lt;p style=&quot;box-sizing: border-box; margin-top: 0px; margin-bottom: 10px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 15px; line-height: 27px; color: rgb(102, 102, 102); text-align: justify; text-wrap: wrap; font-family: 等线; background: rgb(255, 255, 255);&quot;&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px; font-family: 宋体;&quot;&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-family: Arial;&quot;&gt;2.&lt;/span&gt;没有对运动员损伤检测的结果进行进一步的结果分析， 从而导致无法根据检测&lt;/span&gt;&lt;/p&gt;&lt;p style=&quot;box-sizing: border-box; margin-top: 0px; margin-bottom: 10px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 15px; line-height: 27px; color: rgb(102, 102, 102); text-align: justify; text-wrap: wrap; font-family: 等线; background: rgb(255, 255, 255);&quot;&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px; font-family: 宋体;&quot;&gt;结果进行详细判断。&lt;/span&gt;&lt;/p&gt;&lt;p style=&quot;box-sizing: border-box; margin-top: 0px; margin-bottom: 10px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 15px; line-height: 27px; color: rgb(102, 102, 102); text-align: justify; text-wrap: wrap; font-family: 等线; background: rgb(255, 255, 255);&quot;&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px; font-family: 宋体;&quot;&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-family: Arial;&quot;&gt;3.&lt;/span&gt;进行损伤检测时，由于没有进行更详细的检测， 从而导致检测结果不精准&lt;/span&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px; font-family: Arial;&quot;&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-family: 宋体;&quot;&gt;。&lt;/span&gt;&lt;/span&gt;&lt;/p&gt;&lt;h2 style=&quot;box-sizing: border-box; margin: 17px -30px; padding: 10px 0px 10px 40px; outline: 0px; border-width: 0px 0px 0px 4px; border-top-style: initial; border-right-style: initial; border-bottom-style: initial; border-left-style: solid; border-top-color: initial; border-right-color: initial; border-bottom-color: initial; border-left-color: rgb(70, 147, 233); border-image: initial; vertical-align: baseline; font-size: 21px; background: rgb(245, 245, 245); color: rgb(68, 68, 68); text-wrap: wrap; break-after: avoid; text-align: justify; line-height: 36.9067px; font-family: &amp;quot;等线 Light&amp;quot;;&quot;&gt;&lt;strong style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; color: rgb(102, 102, 102); overflow-wrap: break-word;&quot;&gt;2.2 本发明技术方案&lt;/strong&gt;&lt;/h2&gt;&lt;p style=&quot;box-sizing: border-box; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 15px; line-height: 27px; color: rgb(102, 102, 102); text-align: center; text-wrap: wrap; background-color: rgb(255, 255, 255); font-family: 等线;&quot;&gt;&lt;img src=&quot;http://www.trizmed.cn/zb_users/upload/2024/08/202408291724922135553441.png&quot; alt=&quot;image.png&quot;/&gt;&lt;/p&gt;&lt;p style=&quot;box-sizing: border-box; margin-top: 0px; margin-bottom: 10px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 15px; line-height: 27px; color: rgb(102, 102, 102); text-align: justify; text-wrap: wrap; font-family: 等线; background: rgb(255, 255, 255);&quot;&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px; font-family: 宋体;&quot;&gt;&amp;nbsp;&lt;/span&gt;&lt;/p&gt;&lt;h2 style=&quot;box-sizing: border-box; margin: 17px -30px; padding: 10px 0px 10px 40px; outline: 0px; border-width: 0px 0px 0px 4px; border-top-style: initial; border-right-style: initial; border-bottom-style: initial; border-left-style: solid; border-top-color: initial; border-right-color: initial; border-bottom-color: initial; border-left-color: rgb(70, 147, 233); border-image: initial; vertical-align: baseline; font-size: 21px; background: rgb(245, 245, 245); color: rgb(68, 68, 68); text-wrap: wrap; break-after: avoid; text-align: justify; line-height: 36.9067px; font-family: &amp;quot;等线 Light&amp;quot;;&quot;&gt;&lt;strong style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; color: rgb(102, 102, 102); overflow-wrap: break-word;&quot;&gt;2.3 创新点&lt;/strong&gt;&lt;/h2&gt;&lt;p style=&quot;box-sizing: border-box; margin-top: 0px; margin-bottom: 21px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 15px; line-height: 27px; color: rgb(102, 102, 102); text-align: justify; text-wrap: wrap; background-color: rgb(255, 255, 255); font-family: 等线;&quot;&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px;&quot;&gt;利用检测仪器对该运动员进行损伤检测，然后将该运动员的损伤检测数据同该运动员的标准检测数据进行对比比较，进而根据比较结果判该运动员的损伤检测是否合格，如果损伤检测不合格，再对损伤检测数据进行评估，制定训练计划和康复计划。&lt;/span&gt;&lt;/p&gt;&lt;h2 style=&quot;box-sizing: border-box; margin: 17px -30px; padding: 10px 0px 10px 40px; outline: 0px; border-width: 0px 0px 0px 4px; border-top-style: initial; border-right-style: initial; border-bottom-style: initial; border-left-style: solid; border-top-color: initial; border-right-color: initial; border-bottom-color: initial; border-left-color: rgb(70, 147, 233); border-image: initial; vertical-align: baseline; font-size: 21px; background: rgb(245, 245, 245); color: rgb(68, 68, 68); text-wrap: wrap; break-after: avoid; text-align: justify; line-height: 36.9067px; font-family: &amp;quot;等线 Light&amp;quot;;&quot;&gt;&lt;strong style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; color: rgb(102, 102, 102); overflow-wrap: break-word;&quot;&gt;2.4 本发明用到的工具和方法&lt;/strong&gt;&lt;/h2&gt;&lt;h3 style=&quot;box-sizing: border-box; margin: 17px -30px; padding: 7px 0px 6px 40px; outline: 0px; border-width: 0px 0px 0px 4px; border-top-style: initial; border-right-style: initial; border-bottom-style: initial; border-left-style: solid; border-top-color: initial; border-right-color: initial; border-bottom-color: initial; border-left-color: rgb(255, 174, 73); border-image: initial; vertical-align: baseline; font-size: 21px; color: rgb(85, 85, 85); background: rgb(245, 245, 245); text-wrap: wrap; break-after: avoid; text-align: justify; line-height: 36.9067px; font-family: 等线;&quot;&gt;&lt;strong style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; color: rgb(102, 102, 102); overflow-wrap: break-word;&quot;&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px;&quot;&gt;2.4.1 运动员检测补偿系数&lt;/span&gt;&lt;/strong&gt;&lt;/h3&gt;&lt;h2 style=&quot;box-sizing: border-box; margin: 17px -30px; padding: 10px 0px 10px 40px; outline: 0px; border-width: 0px 0px 0px 4px; border-top-style: initial; border-right-style: initial; border-bottom-style: initial; border-left-style: solid; border-top-color: initial; border-right-color: initial; border-bottom-color: initial; border-left-color: rgb(70, 147, 233); border-image: initial; vertical-align: baseline; font-size: 21px; background: rgb(245, 245, 245); color: rgb(68, 68, 68); text-wrap: wrap; break-after: avoid; text-align: justify; line-height: 36.9067px; font-family: &amp;quot;等线 Light&amp;quot;;&quot;&gt;&lt;strong style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; color: rgb(102, 102, 102); overflow-wrap: break-word;&quot;&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px;&quot;&gt;这个概念在权利要求中被提及，并包含了一些看似复杂的数学公式。按照传统的学术派专利写作方法，研究人员往往会在数年的研究后取得成果，并在发表SCI期刊文章前申请专利。这种创新方式效率相对较低。&lt;/span&gt;&lt;/strong&gt;&lt;/h2&gt;&lt;h2 style=&quot;box-sizing: border-box; margin: 17px -30px; padding: 10px 0px 10px 40px; outline: 0px; border-width: 0px 0px 0px 4px; border-top-style: initial; border-right-style: initial; border-bottom-style: initial; border-left-style: solid; border-top-color: initial; border-right-color: initial; border-bottom-color: initial; border-left-color: rgb(70, 147, 233); border-image: initial; vertical-align: baseline; font-size: 21px; background: rgb(245, 245, 245); color: rgb(68, 68, 68); text-wrap: wrap; break-after: avoid; text-align: justify; line-height: 36.9067px; font-family: &amp;quot;等线 Light&amp;quot;;&quot;&gt;&lt;strong style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; color: rgb(102, 102, 102); overflow-wrap: break-word;&quot;&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px;&quot;&gt;从专利的角度看，除非发明人对这些公式进行了实质性的改进，否则在专利中列出这些公式并非必需。专利申请应专注于清晰描述创新点。如果必须涉及现有的方法或公式，可以通过引用相关文献来说明，而无需在专利文本中详细展示公式。总的来说，申请专利时重要的是了解公式的应用目的，而不必深入掌握其数学本质。&lt;/span&gt;&lt;/strong&gt;&lt;/h2&gt;&lt;h2 style=&quot;box-sizing: border-box; margin: 17px -30px; padding: 10px 0px 10px 40px; outline: 0px; border-width: 0px 0px 0px 4px; border-top-style: initial; border-right-style: initial; border-bottom-style: initial; border-left-style: solid; border-top-color: initial; border-right-color: initial; border-bottom-color: initial; border-left-color: rgb(70, 147, 233); border-image: initial; vertical-align: baseline; font-size: 21px; background: rgb(245, 245, 245); color: rgb(68, 68, 68); text-wrap: wrap; break-after: avoid; text-align: justify; line-height: 36.9067px; font-family: &amp;quot;等线 Light&amp;quot;;&quot;&gt;&lt;strong style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; color: rgb(102, 102, 102); overflow-wrap: break-word;&quot;&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px;&quot;&gt;2.5 结合创新工具的思考&lt;/span&gt;&lt;/strong&gt;&lt;/h2&gt;&lt;h2 style=&quot;box-sizing: border-box; margin: 17px -30px; padding: 10px 0px 10px 40px; outline: 0px; border-width: 0px 0px 0px 4px; border-top-style: initial; border-right-style: initial; border-bottom-style: initial; border-left-style: solid; border-top-color: initial; border-right-color: initial; border-bottom-color: initial; border-left-color: rgb(70, 147, 233); border-image: initial; vertical-align: baseline; font-size: 21px; background: rgb(245, 245, 245); color: rgb(68, 68, 68); text-wrap: wrap; break-after: avoid; text-align: justify; line-height: 36.9067px; font-family: &amp;quot;等线 Light&amp;quot;;&quot;&gt;&lt;strong style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; color: rgb(102, 102, 102); overflow-wrap: break-word;&quot;&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px;&quot;&gt;现有专利中使用固定的标准检测数据与运动员的损伤检测数据进行比较。这种方法缺乏灵活性，未能考虑到运动员个体在不同时间点的生理状态变化。比如，一个运动员的身体状态可能因训练周期、年龄、健康状况等多种因素而变化。&lt;/span&gt;&lt;/strong&gt;&lt;/h2&gt;&lt;h2 style=&quot;box-sizing: border-box; margin: 17px -30px; padding: 10px 0px 10px 40px; outline: 0px; border-width: 0px 0px 0px 4px; border-top-style: initial; border-right-style: initial; border-bottom-style: initial; border-left-style: solid; border-top-color: initial; border-right-color: initial; border-bottom-color: initial; border-left-color: rgb(70, 147, 233); border-image: initial; vertical-align: baseline; font-size: 21px; background: rgb(245, 245, 245); color: rgb(68, 68, 68); text-wrap: wrap; break-after: avoid; text-align: justify; line-height: 36.9067px; font-family: &amp;quot;等线 Light&amp;quot;;&quot;&gt;&lt;strong style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; color: rgb(102, 102, 102); overflow-wrap: break-word;&quot;&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px;&quot;&gt;基于此，可以利用TRIZ 40发明原理中的第15条——动态特性对其进行创新，具体来说：设计一个动态调整的检测系统，能够根据运动员的实时表现和历史伤病记录自动调整检测标准和频率。例如，对于康复训练早期的运动员，系统会自动增加检测频率和敏感度，以实现更加精细化的康复训练。&lt;/span&gt;&lt;/strong&gt;&lt;/h2&gt;&lt;p&gt;&lt;br/&gt;&lt;/p&gt;</description><pubDate>Thu, 29 Aug 2024 16:44:40 +0800</pubDate></item><item><title>第三期-VR防眩晕训练器眩晕训练程度动态模拟系统及方法</title><link>http://www.trizmed.cn/jscxd/10.html</link><description>&lt;h1 style=&quot;box-sizing: border-box; margin: 23px -30px 22px; padding: 10px 0px 10px 40px; outline: 0px; border-width: 0px 0px 0px 4px; border-top-style: initial; border-right-style: initial; border-bottom-style: initial; border-left-style: solid; border-top-color: initial; border-right-color: initial; border-bottom-color: initial; border-left-color: rgb(193, 30, 34); border-image: initial; vertical-align: baseline; font-size: 29px; background: rgb(245, 245, 245); color: rgb(68, 68, 68); text-wrap: wrap; break-after: avoid; text-align: justify; line-height: 70.4px; font-family: 等线;&quot;&gt;&lt;strong style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; color: rgb(102, 102, 102); overflow-wrap: break-word;&quot;&gt;1.&amp;nbsp;&lt;/strong&gt;&lt;strong style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; color: rgb(102, 102, 102); overflow-wrap: break-word;&quot;&gt;VR防眩晕训练器眩晕训练程度动态模拟系统及方法&lt;/strong&gt;&lt;/h1&gt;&lt;p style=&quot;box-sizing: border-box; margin-top: 0px; margin-bottom: 10px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 15px; line-height: 27px; color: rgb(102, 102, 102); text-align: justify; text-wrap: wrap; font-family: 等线; background: rgb(255, 255, 255);&quot;&gt;&lt;strong style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; overflow-wrap: break-word;&quot;&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px; font-family: 宋体;&quot;&gt;申请人&lt;/span&gt;&lt;/strong&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px; font-family: 宋体;&quot;&gt;：中国人民解放军总医院&lt;/span&gt;&lt;/p&gt;&lt;p style=&quot;box-sizing: border-box; margin-top: 0px; margin-bottom: 10px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 15px; line-height: 27px; color: rgb(102, 102, 102); text-align: justify; text-wrap: wrap; font-family: 等线; background: rgb(255, 255, 255);&quot;&gt;&lt;strong style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; overflow-wrap: break-word;&quot;&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px; font-family: 宋体;&quot;&gt;申请号&lt;/span&gt;&lt;/strong&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px; font-family: 宋体;&quot;&gt;：&lt;/span&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px; font-family: 宋体;&quot;&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-family: Arial;&quot;&gt;202311388529.0&lt;/span&gt;&lt;/span&gt;&lt;/p&gt;&lt;p style=&quot;box-sizing: border-box; margin-top: 0px; margin-bottom: 10px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 15px; line-height: 27px; color: rgb(102, 102, 102); text-align: justify; text-wrap: wrap; font-family: 等线; background: rgb(255, 255, 255);&quot;&gt;&lt;strong style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; overflow-wrap: break-word;&quot;&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px; font-family: 宋体;&quot;&gt;申请日&lt;/span&gt;&lt;/strong&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px; font-family: 宋体;&quot;&gt;：&lt;/span&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px; font-family: 宋体;&quot;&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-family: Arial;&quot;&gt;2023.10.25&lt;/span&gt;&lt;/span&gt;&lt;/p&gt;&lt;h2 style=&quot;box-sizing: border-box; margin: 17px -30px; padding: 10px 0px 10px 40px; outline: 0px; border-width: 0px 0px 0px 4px; border-top-style: initial; border-right-style: initial; border-bottom-style: initial; border-left-style: solid; border-top-color: initial; border-right-color: initial; border-bottom-color: initial; border-left-color: rgb(70, 147, 233); border-image: initial; vertical-align: baseline; font-size: 21px; background: rgb(245, 245, 245); color: rgb(68, 68, 68); text-wrap: wrap; break-after: avoid; text-align: justify; line-height: 36.9067px; font-family: &amp;quot;等线 Light&amp;quot;;&quot;&gt;&lt;strong style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; color: rgb(102, 102, 102); overflow-wrap: break-word;&quot;&gt;1.1 发明问题&lt;/strong&gt;&lt;/h2&gt;&lt;p style=&quot;box-sizing: border-box; margin-top: 0px; margin-bottom: 10px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 15px; line-height: 27px; color: rgb(102, 102, 102); text-align: justify; text-wrap: wrap; font-family: 等线; background: rgb(255, 255, 255);&quot;&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px; font-family: 宋体;&quot;&gt;防眩晕训练器是一种用于训练防眩晕技能的器材&lt;/span&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px; font-family: Arial;&quot;&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-family: 宋体;&quot;&gt;。&lt;/span&gt;&lt;/span&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px; font-family: 宋体;&quot;&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-family: Arial;&quot;&gt;VR&lt;/span&gt;（虚拟现实）防眩晕训练器的眩晕训练程度动态模拟系统是一种利用&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-family: Arial;&quot;&gt;VR&lt;/span&gt;技术来帮助用户逐步适应可能引起眩晕的虚拟环境的设备。&lt;/span&gt;&lt;/p&gt;&lt;p style=&quot;box-sizing: border-box; margin-top: 0px; margin-bottom: 10px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 15px; line-height: 27px; color: rgb(102, 102, 102); text-align: justify; text-wrap: wrap; font-family: 等线; background: rgb(255, 255, 255);&quot;&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px; font-family: 宋体;&quot;&gt;当完成一次&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-family: Arial;&quot;&gt;VR&lt;/span&gt;防眩晕训练以后会产生一系列的训练数据，如适应性训练师数据、平衡感训练数据、协调性训练数据、身体素质训练数据。如何根据这些数据对训练者的眩晕程度进行准确评估是个亟待解决的问题。&lt;/span&gt;&lt;/p&gt;&lt;h2 style=&quot;box-sizing: border-box; margin: 17px -30px; padding: 10px 0px 10px 40px; outline: 0px; border-width: 0px 0px 0px 4px; border-top-style: initial; border-right-style: initial; border-bottom-style: initial; border-left-style: solid; border-top-color: initial; border-right-color: initial; border-bottom-color: initial; border-left-color: rgb(70, 147, 233); border-image: initial; vertical-align: baseline; font-size: 21px; background: rgb(245, 245, 245); color: rgb(68, 68, 68); text-wrap: wrap; break-after: avoid; text-align: justify; line-height: 36.9067px; font-family: &amp;quot;等线 Light&amp;quot;;&quot;&gt;&lt;strong style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; color: rgb(102, 102, 102); overflow-wrap: break-word;&quot;&gt;1.2 本发明技术方案&lt;/strong&gt;&lt;/h2&gt;&lt;p style=&quot;box-sizing: border-box; margin-top: 0px; margin-right: 16px; margin-bottom: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 15px; line-height: 27px; color: rgb(102, 102, 102); text-align: center; text-wrap: wrap; font-family: 等线; background: rgb(255, 255, 255);&quot;&gt;&lt;img src=&quot;http://www.trizmed.cn/zb_users/upload/2024/08/202408291724922110104630.png&quot; alt=&quot;image.png&quot;/&gt;&lt;/p&gt;&lt;h2 style=&quot;box-sizing: border-box; margin: 17px -30px; padding: 10px 0px 10px 40px; outline: 0px; border-width: 0px 0px 0px 4px; border-top-style: initial; border-right-style: initial; border-bottom-style: initial; border-left-style: solid; border-top-color: initial; border-right-color: initial; border-bottom-color: initial; border-left-color: rgb(70, 147, 233); border-image: initial; vertical-align: baseline; font-size: 21px; background: rgb(245, 245, 245); color: rgb(68, 68, 68); text-wrap: wrap; break-after: avoid; text-align: justify; line-height: 36.9067px; font-family: &amp;quot;等线 Light&amp;quot;;&quot;&gt;&lt;strong style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; color: rgb(102, 102, 102); overflow-wrap: break-word;&quot;&gt;1.3 创新点&lt;/strong&gt;&lt;/h2&gt;&lt;p style=&quot;box-sizing: border-box; margin-top: 0px; margin-bottom: 10px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 15px; line-height: 27px; color: rgb(102, 102, 102); text-align: justify; text-wrap: wrap; font-family: 等线; background: rgb(255, 255, 255);&quot;&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px; font-family: 宋体;&quot;&gt;为了解决对训练结果数据优化不佳以及评估不准确的问题。本发明主要有两个创新点。&lt;/span&gt;&lt;/p&gt;&lt;p style=&quot;box-sizing: border-box; margin-top: 0px; margin-bottom: 10px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 15px; line-height: 27px; color: rgb(102, 102, 102); text-align: justify; text-wrap: wrap; font-family: 等线; background: rgb(255, 255, 255);&quot;&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px; font-family: 宋体;&quot;&gt;第一个创新点是通过比较优化曲线数据和标准曲线数据，确定训练评级。具体来说，将训练数据预处理后导入模拟模型架构中处理得到优化曲线数据，同时一份训练数据还对应一份标准曲线数据，然后将优化曲线数据同标准曲线数据进行比对，根据线条的重叠长度确定训练者的评估结果。&lt;/span&gt;&lt;/p&gt;&lt;p style=&quot;box-sizing: border-box; margin-top: 0px; margin-bottom: 10px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 15px; line-height: 27px; color: rgb(102, 102, 102); text-align: justify; text-wrap: wrap; font-family: 等线; background: rgb(255, 255, 255);&quot;&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px; font-family: 宋体;&quot;&gt;第二个创新点是根据梯度递减公式和反向传播机制优化标准曲线数据。&lt;/span&gt;&lt;/p&gt;&lt;p style=&quot;box-sizing: border-box; margin-top: 0px; margin-right: 16px; margin-bottom: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 15px; line-height: 27px; color: rgb(102, 102, 102); text-align: justify; text-wrap: wrap; font-family: 等线; background: rgb(255, 255, 255);&quot;&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px; font-family: 宋体;&quot;&gt;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&lt;/span&gt;&lt;/p&gt;&lt;h2 style=&quot;box-sizing: border-box; margin: 17px -30px; padding: 10px 0px 10px 40px; outline: 0px; border-width: 0px 0px 0px 4px; border-top-style: initial; border-right-style: initial; border-bottom-style: initial; border-left-style: solid; border-top-color: initial; border-right-color: initial; border-bottom-color: initial; border-left-color: rgb(70, 147, 233); border-image: initial; vertical-align: baseline; font-size: 21px; background: rgb(245, 245, 245); color: rgb(68, 68, 68); text-wrap: wrap; break-after: avoid; text-align: justify; line-height: 36.9067px; font-family: &amp;quot;等线 Light&amp;quot;;&quot;&gt;&lt;strong style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; color: rgb(102, 102, 102); overflow-wrap: break-word;&quot;&gt;1.&lt;/strong&gt;&lt;strong style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; color: rgb(102, 102, 102); overflow-wrap: break-word;&quot;&gt;4&lt;/strong&gt;&lt;strong style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; color: rgb(102, 102, 102); overflow-wrap: break-word;&quot;&gt;&amp;nbsp;结合创新工具的思考&lt;/strong&gt;&lt;/h2&gt;&lt;p style=&quot;box-sizing: border-box; margin-top: 0px; margin-bottom: 21px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 15px; line-height: 27px; color: rgb(102, 102, 102); text-align: justify; text-wrap: wrap; background-color: rgb(255, 255, 255); font-family: 等线;&quot;&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px; font-family: 宋体;&quot;&gt;本发明专利读起来比较费力，里面提到的&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-family: Arial;&quot;&gt;SCDM&lt;/span&gt;几何创建法同&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-family: Arial;&quot;&gt;Google&lt;/span&gt;出来的概念也不是同一个，可能需要具体研究这个项目的人才能真正理解本发明背后的实现逻辑&lt;/span&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px; font-family: Arial;&quot;&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-family: 宋体;&quot;&gt;。&lt;/span&gt;&lt;/span&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px; font-family: 宋体;&quot;&gt;该发明专利仍然是有启发意义的。这些意义体现在以下几个方面：&lt;/span&gt;&lt;/p&gt;&lt;p style=&quot;box-sizing: border-box; margin-top: 0px; margin-bottom: 21px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 15px; line-height: 27px; color: rgb(102, 102, 102); text-align: justify; text-wrap: wrap; background-color: rgb(255, 255, 255); font-family: 等线;&quot;&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px; font-family: 宋体;&quot;&gt;1)&amp;nbsp;&lt;/span&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px; font-family: 宋体;&quot;&gt;介绍了一个新的发明问题，即如何利用训练数据来进行眩晕程度评级，没有发明&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-family: Arial;&quot;&gt;idea&lt;/span&gt;的人可以借鉴，但是如果对于非该领域的人来说，如果想要彻底弄懂这个技术问题，还要读更多的同类型专利或文章。&lt;/span&gt;&lt;/p&gt;&lt;p style=&quot;box-sizing: border-box; margin-top: 0px; margin-bottom: 21px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 15px; line-height: 27px; color: rgb(102, 102, 102); text-align: justify; text-wrap: wrap; background-color: rgb(255, 255, 255); font-family: 等线;&quot;&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px; font-family: 宋体; background: #FFFF00;&quot;&gt;2)&amp;nbsp;&lt;/span&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px; font-family: 宋体; background: #FFFF00;&quot;&gt;提供了一个处理问题的简单范式，即拿标准数据同待评测数据进行比较，得到评估效果。这种处理问题的范式是可以应用到其他问题中的。&lt;/span&gt;&lt;/p&gt;&lt;p style=&quot;box-sizing: border-box; margin-top: 0px; margin-bottom: 21px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 15px; line-height: 27px; color: rgb(102, 102, 102); text-align: justify; text-wrap: wrap; background-color: rgb(255, 255, 255); font-family: 等线;&quot;&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px; font-family: 宋体;&quot;&gt;3)&amp;nbsp;&lt;/span&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px; font-family: 宋体;&quot;&gt;这里标准数据的获取是通过算法优化的方式获得的，而人工智能中的标准数据大多是人工获得的，这也就是业界常说的有多少人工就有多少智能。人工智能中的标准数据被称作标签，比如对于猫狗图像二分类任务来说，一张图像是猫还是狗，这叫做标签；比如对于&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-family: Arial;&quot;&gt;car/truck/pedestiran&lt;/span&gt;等目标的检测任务来说，这些目标在图像中的坐标叫做标签（可以以图像左上角为原点，横向向右为&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-family: Arial;&quot;&gt;x&lt;/span&gt;轴正方向，纵向向下为&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-family: Arial;&quot;&gt;y&lt;/span&gt;轴正方向建立坐标系，通过画矩形框来框柱上述目标，对应矩形框的坐标就是标签）。那么能否以优化的方式获得人工智能数据中的标签呢？&lt;/span&gt;&lt;/p&gt;&lt;p style=&quot;box-sizing: border-box; margin-top: 0px; margin-bottom: 21px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 15px; line-height: 27px; color: rgb(102, 102, 102); text-align: justify; text-wrap: wrap; background-color: rgb(255, 255, 255); font-family: 等线;&quot;&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px; font-family: 宋体;&quot;&gt;4)&amp;nbsp;&lt;/span&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px; font-family: 宋体;&quot;&gt;本发明问题中也是有各式各样的数据的，比如眩晕训练数据中的适应性数据、平衡感数据、协调性数据、身体素质训练数据，并且我们知道机器学习和或者深度学习都是从数据中学习规律的，那么能否直接用机器学习方法或者深度学习方法直接从这些数据中获得训练的评估结果呢？&lt;/span&gt;&lt;/p&gt;&lt;p&gt;&lt;br/&gt;&lt;/p&gt;</description><pubDate>Thu, 29 Aug 2024 16:44:17 +0800</pubDate></item><item><title>第二期-左心室磁共振影像智能分类方法、装置、设备和介质</title><link>http://www.trizmed.cn/jscxd/9.html</link><description>&lt;h1 style=&quot;box-sizing: border-box; margin: 23px -30px 22px; padding: 10px 0px 10px 40px; outline: 0px; border-width: 0px 0px 0px 4px; border-top-style: initial; border-right-style: initial; border-bottom-style: initial; border-left-style: solid; border-top-color: initial; border-right-color: initial; border-bottom-color: initial; border-left-color: rgb(193, 30, 34); border-image: initial; vertical-align: baseline; font-size: 29px; background: rgb(245, 245, 245); color: rgb(68, 68, 68); text-wrap: wrap; break-after: avoid; text-align: justify; line-height: 70.4px; font-family: 等线;&quot;&gt;&lt;strong style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; color: rgb(102, 102, 102); overflow-wrap: break-word;&quot;&gt;1.&amp;nbsp;&lt;/strong&gt;&lt;strong style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; color: rgb(102, 102, 102); overflow-wrap: break-word;&quot;&gt;左心室磁共振影像智能分类方法、装置、设备和介质&lt;/strong&gt;&lt;/h1&gt;&lt;p style=&quot;box-sizing: border-box; margin-top: 0px; margin-bottom: 10px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 15px; line-height: 27px; color: rgb(102, 102, 102); text-align: justify; text-wrap: wrap; font-family: 等线; background: rgb(255, 255, 255);&quot;&gt;&lt;strong style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; overflow-wrap: break-word;&quot;&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px; font-family: 宋体;&quot;&gt;申请人&lt;/span&gt;&lt;/strong&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px; font-family: 宋体;&quot;&gt;：中国人民解放军总医院&lt;/span&gt;&lt;/p&gt;&lt;p style=&quot;box-sizing: border-box; margin-top: 0px; margin-bottom: 10px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 15px; line-height: 27px; color: rgb(102, 102, 102); text-align: justify; text-wrap: wrap; font-family: 等线; background: rgb(255, 255, 255);&quot;&gt;&lt;strong style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; overflow-wrap: break-word;&quot;&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px; font-family: 宋体;&quot;&gt;申请号&lt;/span&gt;&lt;/strong&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px; font-family: 宋体;&quot;&gt;：&lt;/span&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px; font-family: 宋体;&quot;&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-family: Arial;&quot;&gt;202110077126.9&lt;/span&gt;&lt;/span&gt;&lt;/p&gt;&lt;p style=&quot;box-sizing: border-box; margin-top: 0px; margin-bottom: 10px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 15px; line-height: 27px; color: rgb(102, 102, 102); text-align: justify; text-wrap: wrap; font-family: 等线; background: rgb(255, 255, 255);&quot;&gt;&lt;strong style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; overflow-wrap: break-word;&quot;&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px; font-family: 宋体;&quot;&gt;申请日&lt;/span&gt;&lt;/strong&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px; font-family: 宋体;&quot;&gt;：&lt;/span&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px; font-family: 宋体;&quot;&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-family: Arial;&quot;&gt;2021.01.20&lt;/span&gt;&lt;/span&gt;&lt;/p&gt;&lt;h2 style=&quot;box-sizing: border-box; margin: 17px -30px; padding: 10px 0px 10px 40px; outline: 0px; border-width: 0px 0px 0px 4px; border-top-style: initial; border-right-style: initial; border-bottom-style: initial; border-left-style: solid; border-top-color: initial; border-right-color: initial; border-bottom-color: initial; border-left-color: rgb(70, 147, 233); border-image: initial; vertical-align: baseline; font-size: 21px; background: rgb(245, 245, 245); color: rgb(68, 68, 68); text-wrap: wrap; break-after: avoid; text-align: justify; line-height: 36.9067px; font-family: &amp;quot;等线 Light&amp;quot;;&quot;&gt;&lt;strong style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; color: rgb(102, 102, 102); overflow-wrap: break-word;&quot;&gt;1.1 发明问题&lt;/strong&gt;&lt;/h2&gt;&lt;p style=&quot;box-sizing: border-box; margin-top: 0px; margin-bottom: 10px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 15px; line-height: 27px; color: rgb(102, 102, 102); text-align: justify; text-wrap: wrap; font-family: 等线; background: rgb(255, 255, 255);&quot;&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px; font-family: 宋体;&quot;&gt;随着深度学习技术的快速发展，医学影像处理有望提供低成本，高效，准确的辅助医疗手段。在心脏疾病领域，心脏磁共振凭借其高分辨率，高软组织对比度以及多序列成像的优势，成为心肌病分类的重要标准。&lt;/span&gt;&lt;/p&gt;&lt;p style=&quot;box-sizing: border-box; margin-top: 0px; margin-bottom: 10px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 15px; line-height: 27px; color: rgb(102, 102, 102); text-align: justify; text-wrap: wrap; font-family: 等线; background: rgb(255, 255, 255);&quot;&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px; font-family: 宋体;&quot;&gt;相关技术中，通常需要对核磁影像进行分割，心功能指标提取等复杂流程。然而，在整个流程中，需要大量的人工操作，导致效率较低，引入误差，准确率偏低，亟待解决&lt;/span&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px; font-family: Arial;&quot;&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-family: 宋体;&quot;&gt;。&lt;/span&gt;&lt;/span&gt;&lt;/p&gt;&lt;h2 style=&quot;box-sizing: border-box; margin: 17px -30px; padding: 10px 0px 10px 40px; outline: 0px; border-width: 0px 0px 0px 4px; border-top-style: initial; border-right-style: initial; border-bottom-style: initial; border-left-style: solid; border-top-color: initial; border-right-color: initial; border-bottom-color: initial; border-left-color: rgb(70, 147, 233); border-image: initial; vertical-align: baseline; font-size: 21px; background: rgb(245, 245, 245); color: rgb(68, 68, 68); text-wrap: wrap; break-after: avoid; text-align: justify; line-height: 36.9067px; font-family: &amp;quot;等线 Light&amp;quot;;&quot;&gt;&lt;strong style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; color: rgb(102, 102, 102); overflow-wrap: break-word;&quot;&gt;1.2 本发明技术方案&lt;/strong&gt;&lt;/h2&gt;&lt;p style=&quot;box-sizing: border-box; margin-top: 0px; margin-right: 16px; margin-bottom: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 15px; line-height: 27px; color: rgb(102, 102, 102); text-align: center; text-wrap: wrap; font-family: 等线; background: rgb(255, 255, 255);&quot;&gt;&lt;a class=&quot;view-image&quot; href=&quot;https://www.kyjszj.com/zb_users/upload/2024/08/20240829092028172489442871034.jpeg&quot; title=&quot;17219733033311.png&quot; alt=&quot;1715387784041.png&quot; data-fancybox=&quot;gallery&quot; style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; background: transparent; color: rgb(50, 165, 231); vertical-align: baseline; text-decoration-line: none; font-family: &amp;quot;Open Sans&amp;quot;, &amp;quot;Hiragino Sans GB&amp;quot;, &amp;quot;Microsoft YaHei&amp;quot;, &amp;quot;WenQuanYi Micro Hei&amp;quot;, Arial, Verdana, Tahoma, sans-serif; -webkit-tap-highlight-color: rgba(255, 0, 0, 0); cursor: url(&amp;quot;https://www.kyjszj.com/zb_users/plugin/Jz52_Pointer/Pointer/materialdesign/link.png&amp;quot;), auto !important;&quot;&gt;&lt;/a&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px; font-family: 宋体;&quot;&gt;&amp;nbsp;&lt;img src=&quot;http://www.trizmed.cn/zb_users/upload/2024/08/202408291724922086287410.png&quot; alt=&quot;image.png&quot;/&gt;&lt;/span&gt;&lt;/p&gt;&lt;p style=&quot;box-sizing: border-box; margin-top: 0px; margin-bottom: 10px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 15px; line-height: 27px; color: rgb(102, 102, 102); text-align: justify; text-wrap: wrap; font-family: 等线; background: rgb(255, 255, 255);&quot;&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px; font-family: 宋体;&quot;&gt;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&lt;/span&gt;&lt;/p&gt;&lt;h2 style=&quot;box-sizing: border-box; margin: 17px -30px; padding: 10px 0px 10px 40px; outline: 0px; border-width: 0px 0px 0px 4px; border-top-style: initial; border-right-style: initial; border-bottom-style: initial; border-left-style: solid; border-top-color: initial; border-right-color: initial; border-bottom-color: initial; border-left-color: rgb(70, 147, 233); border-image: initial; vertical-align: baseline; font-size: 21px; background: rgb(245, 245, 245); color: rgb(68, 68, 68); text-wrap: wrap; break-after: avoid; text-align: justify; line-height: 36.9067px; font-family: &amp;quot;等线 Light&amp;quot;;&quot;&gt;&lt;strong style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; color: rgb(102, 102, 102); overflow-wrap: break-word;&quot;&gt;1.3 创新点&lt;/strong&gt;&lt;/h2&gt;&lt;p style=&quot;box-sizing: border-box; margin-top: 0px; margin-bottom: 10px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 15px; line-height: 27px; color: rgb(102, 102, 102); text-align: justify; text-wrap: wrap; font-family: 等线; background: rgb(255, 255, 255);&quot;&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px; font-family: 宋体;&quot;&gt;本发明&lt;/span&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px; font-family: 宋体;&quot;&gt;使用了多模态数据融合来预测心脏左心室图像的分类概率，具体来说融合了用户的症状信息、病史信息、生理指标信息、心脏共振影像视频信息&lt;/span&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px; font-family: 宋体;&quot;&gt;。&lt;/span&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px; font-family: 宋体;&quot;&gt;首先将用户的症状信息、病史信息、生理指标信息送输入第一分析模型，得到第一分析结果；然后把心脏共振影像视频信息输入到第二分析模型，得到第二分析结果；然后把第一分析结果和第二分析结果进行拼接，作为第三分析结果，然后把第三分析结果输入第三分析模型，进而得到目标用户的左心室图像的分类概率值。&lt;/span&gt;&lt;/p&gt;&lt;p style=&quot;box-sizing: border-box; margin-top: 0px; margin-right: 16px; margin-bottom: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 15px; line-height: 27px; color: rgb(102, 102, 102); text-align: justify; text-wrap: wrap; font-family: 等线; background: rgb(255, 255, 255);&quot;&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px; font-family: 宋体;&quot;&gt;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp;&lt;/span&gt;&lt;/p&gt;&lt;h2 style=&quot;box-sizing: border-box; margin: 17px -30px; padding: 10px 0px 10px 40px; outline: 0px; border-width: 0px 0px 0px 4px; border-top-style: initial; border-right-style: initial; border-bottom-style: initial; border-left-style: solid; border-top-color: initial; border-right-color: initial; border-bottom-color: initial; border-left-color: rgb(70, 147, 233); border-image: initial; vertical-align: baseline; font-size: 21px; background: rgb(245, 245, 245); color: rgb(68, 68, 68); text-wrap: wrap; break-after: avoid; text-align: justify; line-height: 36.9067px; font-family: &amp;quot;等线 Light&amp;quot;;&quot;&gt;&lt;strong style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; color: rgb(102, 102, 102); overflow-wrap: break-word;&quot;&gt;1.4 本发明&lt;/strong&gt;&lt;strong style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; color: rgb(102, 102, 102); overflow-wrap: break-word;&quot;&gt;涉及到&lt;/strong&gt;&lt;strong style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; color: rgb(102, 102, 102); overflow-wrap: break-word;&quot;&gt;的工具和方法：&lt;/strong&gt;&lt;/h2&gt;&lt;h3 style=&quot;box-sizing: border-box; margin: 17px -30px; padding: 7px 0px 6px 40px; outline: 0px; border-width: 0px 0px 0px 4px; border-top-style: initial; border-right-style: initial; border-bottom-style: initial; border-left-style: solid; border-top-color: initial; border-right-color: initial; border-bottom-color: initial; border-left-color: rgb(255, 174, 73); border-image: initial; vertical-align: baseline; font-size: 21px; color: rgb(85, 85, 85); background: rgb(245, 245, 245); text-wrap: wrap; break-after: avoid; text-align: justify; line-height: 36.9067px; font-family: 等线;&quot;&gt;&lt;strong style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; color: rgb(102, 102, 102); overflow-wrap: break-word;&quot;&gt;1.4.1&amp;nbsp;&lt;/strong&gt;&lt;strong style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; color: rgb(102, 102, 102); overflow-wrap: break-word;&quot;&gt;word2vector&lt;/strong&gt;&lt;strong style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; color: rgb(102, 102, 102); overflow-wrap: break-word;&quot;&gt;:&lt;/strong&gt;&lt;/h3&gt;&lt;p style=&quot;box-sizing: border-box; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 15px; line-height: 27px; color: rgb(102, 102, 102); text-align: justify; text-wrap: wrap; background-color: rgb(255, 255, 255); font-family: 等线;&quot;&gt;&amp;nbsp;&lt;/p&gt;&lt;p style=&quot;box-sizing: border-box; margin-top: 0px; margin-bottom: 21px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 15px; line-height: 27px; color: rgb(102, 102, 102); text-align: justify; text-wrap: wrap; background-color: rgb(255, 255, 255); font-family: 等线;&quot;&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px;&quot;&gt;Word2Vec是一种广泛使用的自然语言处理（NLP）技术，用于将单词表示为向量，这些向量在多维空间中以一种能够捕捉单词之间语义和语法关系的方式进行编码。这种方法由Google的研究团队在2013年提出，并迅速成为了文本处理任务中的一个基础工具。&lt;/span&gt;&lt;/p&gt;&lt;p style=&quot;box-sizing: border-box; margin-top: 0px; margin-bottom: 21px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 15px; line-height: 27px; color: rgb(102, 102, 102); text-align: justify; text-wrap: wrap; background-color: rgb(255, 255, 255); font-family: 等线;&quot;&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px;&quot;&gt;Word2Vec的向量表示形式使得它可以广泛应用于多种NLP任务，包括：1) 文本相似度计算：通过比较单词或文档向量之间的距离（如余弦相似性），可以评估文本内容的相似度；2) 文本分类和情感分析：将文本转换为向量表示，然后使用机器学习模型进行分类，如判断评论是正面还是负面；3) 机器翻译：单词向量可以用来捕捉语言之间的语义关系，辅助翻译算法更好地理解和转换意义；4) 推荐系统：在推荐系统中，商品描述的文本信息可以转化为向量，通过计算向量相似性来推荐相似商品。&lt;/span&gt;&lt;/p&gt;&lt;h3 style=&quot;box-sizing: border-box; margin: 17px -30px; padding: 7px 0px 6px 40px; outline: 0px; border-width: 0px 0px 0px 4px; border-top-style: initial; border-right-style: initial; border-bottom-style: initial; border-left-style: solid; border-top-color: initial; border-right-color: initial; border-bottom-color: initial; border-left-color: rgb(255, 174, 73); border-image: initial; vertical-align: baseline; font-size: 21px; color: rgb(85, 85, 85); background: rgb(245, 245, 245); text-wrap: wrap; break-after: avoid; text-align: justify; line-height: 36.9067px; font-family: 等线;&quot;&gt;&lt;strong style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; color: rgb(102, 102, 102); overflow-wrap: break-word;&quot;&gt;1.4.2&amp;nbsp;&lt;/strong&gt;&lt;strong style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; color: rgb(102, 102, 102); overflow-wrap: break-word;&quot;&gt;特征拼接层、残差卷积层、全局池化层、位置编码层&lt;/strong&gt;&lt;/h3&gt;&lt;p style=&quot;box-sizing: border-box; margin-top: 0px; margin-bottom: 10px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 15px; line-height: 27px; color: rgb(102, 102, 102); text-align: justify; text-wrap: wrap; background-color: rgb(255, 255, 255); font-family: 等线;&quot;&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px;&quot;&gt;这些概念属于深度学习神经网络中常用的层结构。在撰写专利时，若未对算法进行创新性改进，无需对算法细节进行过度详细的描述。接下来，我们将对深度学习神经网络，简称为深度学习，进行介绍。&lt;/span&gt;&lt;/p&gt;&lt;p style=&quot;box-sizing: border-box; margin-top: 0px; margin-bottom: 10px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 15px; line-height: 27px; color: rgb(102, 102, 102); text-align: justify; text-wrap: wrap; background-color: rgb(255, 255, 255); font-family: 等线;&quot;&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px;&quot;&gt;深度学习是一种人工智能技术，它基于人工神经网络的架构来模拟人脑分析和学习数据的方式。深度学习的“深度”指的是模型的层次结构中包含多层，这些层次可以从数据中学习到不同级别的抽象特征。每一层自动从其前一层学习到的特征中提取更加复杂的高级特征。深度学习模型通常需要大量的数据和计算能力才能有效地训练。&lt;/span&gt;&lt;/p&gt;&lt;p style=&quot;box-sizing: border-box; margin-top: 0px; margin-bottom: 10px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 15px; line-height: 27px; color: rgb(102, 102, 102); text-align: justify; text-wrap: wrap; background-color: rgb(255, 255, 255); font-family: 等线;&quot;&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px;&quot;&gt;深度学习可以用来完成人脸识别、语音识别、自然语言处理、疾病预测等。在这篇专利里，深度学习被用来完成分类任务，用来&lt;/span&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px; font-family: 宋体;&quot;&gt;得到目标用户的左心室图像的分类概率值。&lt;/span&gt;&lt;/p&gt;&lt;h2 style=&quot;box-sizing: border-box; margin: 17px -30px; padding: 10px 0px 10px 40px; outline: 0px; border-width: 0px 0px 0px 4px; border-top-style: initial; border-right-style: initial; border-bottom-style: initial; border-left-style: solid; border-top-color: initial; border-right-color: initial; border-bottom-color: initial; border-left-color: rgb(70, 147, 233); border-image: initial; vertical-align: baseline; font-size: 21px; background: rgb(245, 245, 245); color: rgb(68, 68, 68); text-wrap: wrap; break-after: avoid; text-align: justify; line-height: 36.9067px; font-family: &amp;quot;等线 Light&amp;quot;;&quot;&gt;&lt;strong style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; color: rgb(102, 102, 102); overflow-wrap: break-word;&quot;&gt;1.5 结合创新工具的思考&lt;/strong&gt;&lt;/h2&gt;&lt;p style=&quot;box-sizing: border-box; margin-top: 0px; margin-bottom: 10px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 15px; line-height: 27px; color: rgb(102, 102, 102); text-align: justify; text-wrap: wrap; background-color: rgb(255, 255, 255); font-family: 等线;&quot;&gt;本发明专利中提到的第一、第二等词语是专利中经常遇到的，这是一种上位的描述方式，上位这一概念是相对下位来说的，也可以理解为一般与具体的关系，在课程中对这一概念也有介绍。&lt;/p&gt;&lt;p style=&quot;box-sizing: border-box; margin-top: 0px; margin-bottom: 10px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 15px; line-height: 27px; color: rgb(102, 102, 102); text-align: justify; text-wrap: wrap; background-color: rgb(255, 255, 255); font-family: 等线;&quot;&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px; font-family: Arial;&quot;&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-family: 宋体;&quot;&gt;我们可以考虑利用&lt;/span&gt;TRIZ&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-family: 宋体;&quot;&gt;第&lt;/span&gt;&lt;/span&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px; font-family: 宋体;&quot;&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-family: Arial;&quot;&gt;1&lt;/span&gt;&lt;/span&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px; font-family: Arial;&quot;&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-family: 宋体;&quot;&gt;条发明原理中介绍的&lt;/span&gt;&lt;/span&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px; font-family: 宋体;&quot;&gt;分割&lt;/span&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px; font-family: Arial;&quot;&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-family: 宋体;&quot;&gt;原理&lt;/span&gt;&lt;/span&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px; font-family: 宋体;&quot;&gt;从时间和空间两个角度&lt;/span&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px; font-family: Arial;&quot;&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-family: 宋体;&quot;&gt;来进行创新。&lt;/span&gt;&lt;/span&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px; font-family: 宋体;&quot;&gt;结&lt;/span&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px; font-family: 宋体;&quot;&gt;合&lt;/span&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px; font-family: 宋体;&quot;&gt;分割&lt;/span&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px; font-family: 宋体;&quot;&gt;作用原理，&lt;/span&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px; font-family: 宋体;&quot;&gt;从时间分离角度，可以将数据处理过程分时段进行，如在数据采集阶段先处理症状信息和病史信息，而将影像数据的处理延后，以优化处理速度和准确性。从空间分离角度，可以将影像数据处理与其他数据处理在不同的系统或模块中进行，提高系统的灵活性和可维护性。&lt;/span&gt;&lt;span style=&quot;box-sizing: border-box; margin: 0px; padding: 0px; outline: 0px; border: 0px; vertical-align: baseline; font-size: 16px;&quot;&gt;&amp;nbsp;&lt;/span&gt;&lt;/p&gt;&lt;p&gt;&lt;br/&gt;&lt;/p&gt;</description><pubDate>Thu, 29 Aug 2024 16:42:55 +0800</pubDate></item></channel></rss>