基于MLFCC特征和P-CRNN的心音分类研究  

Research on Heart Sound Classification Based on Mixed Features of MFCC and LFCC and Parallel Convolutional Recurrent Neural Network

作　　者：宋晨翔 张孙杰 刘奥磊 王哲 Chenxiang Song;Sunjie Zhang;Aolei Liu;Zhe Wang(School of Optical-Electrical and Computer Engineering,University of Shanghai for Science and Technology,Shanghai)

机构地区：[1]上海理工大学光电信息与计算机工程学院,上海

出　　处：《建模与仿真》2025年第2期202-212,共11页Modeling and Simulation

基　　金：国家自然科学基金(No.61603255);上海市晨光计划(NO.18CG52)。

摘　　要：心音分类是医学领域中一个重要的诊断任务。但由于数据量不足、宽频带信号表征能力欠缺、心音序列上下文学习存在挑战等问题,该领域依旧存在许多的不足之处。针对以上问题,提出了一种心音分类的新方法:对每段心音信号进行重叠切片,截取成2s的信号作为样本;随后采用改进的梅尔频率倒谱系数(MFCC)与线性频率倒谱系数(LFCC)分别提取心音信号相应频率系数,并分别计算其一阶差分作为融合特征。分类网络使用提出的并行卷积递归神经网络(P-CRNN)方法进行训练。实验表明,相比于其他传统识别方法,所提方法对心音信号分类有明显提高。Heart sound classification is a crucial diagnostic task in the medical field.However,due to the insufficient amount of data,the lack of wideband signal representation ability,and the challenges of contextual learning of heart sound sequences,there are still many shortcomings in this field.To address these issues,a novel method for heart sound classification was proposed.In this method,each signal was segmented into 2-second overlapping slices,and the corresponding frequency coefficients were extracted using improved Mel Frequency Cepstral Coefficients(MFCC)and Linear Frequency Cepstral Coefficients(LFCC).The first-order difference was calculated as fusion features,and the proposed Parallel Convolutional Recurrent Neural Network(P-CRNN)method was trained for classification in comparison to traditional recognition methods.The experiment shows that compared to other traditional recognition methods,the proposed method has a significant improvement in heart sound signal classification.

关 键 词：心音信号 混合特征 分类网络 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]