基于特征选择和降维的心音信号识别的研究  

Research on New Sound Signal Recognition Based on Feature Selection and Dimension Reduction

作　　者：李海霞[1] 朱慧博[1] 张施琴[1] 毛凌锋 LI Haixia;ZHU Huibo;ZHANG Shiqin;MAO Lingfeng(Suqian College,Suqian 223800;University of Science&Technology Beijing,Beijing 100083)

机构地区：[1]宿迁学院,宿迁223800 [2]北京科技大学,北京100083

出　　处：《计算机与数字工程》2022年第5期1131-1135,共5页Computer & Digital Engineering

基　　金：国家自然科学基金项目(编号:61774014);江苏省高校自然科学研究面上项目(编号:19KJB510060);江苏省教育科学“十三五”课题(编号:C-c/2020/01/17,C-c/2018/01/04);江苏省研究生科研与实践创新项目(编号:KYZZ15_0331);江苏省六大人才高峰项目(编号:XCL227)资助。

摘　　要：为提高心音信号识别的高效性,论文将特征选择后的参数进行降维处理,得到计算成本降低且识别率满足需求的心音识别模型。首先使用小波包分解和功率谱估计计算心音参数,接下来使用4种方法进行特征选择,通过比较识别率,找出能较好反映心音本质特征且具有差异性的参数,最后再将特征选择得到的参数分别用主成分分析法和线性判别分析法进行降维处理,并分别计算降维处理后的识别率。210例样本的识别测试结果表明,降维后的信号识别率达到92.9%,证明该方法能在降维的基础上保证足够的识别率。In order to improve the efficiency of heart sound signal recognition,this paper reduces the dimensionality of the parameters after feature selection,and obtains a heart sound recognition model with reduced computation cost and satisfactory recognition rate.Firstly,wavelet packet decomposition and power spectrum estimation are used to calculate the heart sound parameters.Next,four methods are used to select the features.By comparing the recognition rates,the parameters that can better reflect the essential features of the heart sounds can be found,and finally the features can be selected.The parameters are reduced by principal component analysis and linear discriminant analysis,and the recognition rate after dimension reduction is calculated.The recognition test results of 210 samples show that the signal recognition rate after dimension reduction reaches 92.9%,which proves that the method can guarantee sufficient recognition rate based on dimensionality reduction.

关 键 词：心音 特征选择 降维 心音识别 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]