基于APSOC的心音特征提取及分类  

Feature extraction and classification of heart sounds based on APSOC

作　　者：田英杰 杨宏波 汪琴 郭涛 潘家华 王威廉[1] TIAN Ying-jie;YANG Hong-bo;WANG Qin;GUO Tao;PAN Jia-hua;WANG Wei-lian(School of Information Science and Engineering,Yunnan University,Kunming 650500,China;Structural Heart Disease Ward,Fuwai Yunnan Cardiovascular Hospital,Kunming 650102,China)

机构地区：[1]云南大学信息学院,云南昆明650500 [2]云南省阜外心血管病医院结构性心脏病病区,云南昆明650102

出　　处：《计算机工程与设计》2024年第12期3779-3785,共7页Computer Engineering and Design

基　　金：国家自然科学基金项目(81960067);2018云南省重大科技专项基金项目(2018ZF017)。

摘　　要：在云南一些边远山区网络信号弱甚至无信号,为在便携式设备上实现心音分类算法,满足离线式、可移动的需求,提出一种可部署在APSOC平台上的心音分类方法。在PS部分实现心音信号的特征提取,在PL部分实现CNN的卷积层和池化层。使用多通道并行及流水线等方式,实现对系统的硬件加速。实验结果表明,与通用CPU相比,该方法实现了8.91倍的硬件加速,分类准确率仅损失了2%,对心音辅助诊断有实用价值。In some remote mountainous areas of Yunnan,the network signal is weak or no signals even.To implement heart sound classification algorithms on portable devices and meet the offline and mobile needs,a heart sound classification method that could be deployed on the APSOC platform was proposed.The feature extraction of heart sound signals was implemented in the PS section,and convolutional and pooling layers of CNN were implemented in the PL section.Multi-channel parallel and pipeline methods were used to achieve hardware acceleration of the system.Experimental results show that compared to general-purpose CPUs,it achieves 8.91 times hardware acceleration with only 2%classification accuracy lost.Experimental results indicate that the proposed scheme has practical value for assisting in the diagnosis of heart sounds.

关 键 词：全可编程片上系统 心音分类 先天性心脏病 硬件加速 卷积神经网络 梅尔频率倒谱系数 并行计算 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]