基于WST-BiLSTM的心音信号分类研究  

Research on Heart Sound Signal Classification Based on WST-BiLSTM

作　　者：夏文鑫 蒋捷 张大斌[1] 李晖[3] 赖智通 XIA Wenxin;JIANG Jie;ZHANG Dabin;LI Hui;LAI Zhitong(School of Mechanical Engineering,Guizhou University,Guiyang 550025;The Affiliated Hospital of Guizhou Medical University,Guiyang 550025;College of Computer Science and Technology,Guizhou University,Guiyang 550025)

机构地区：[1]贵州大学机械工程学院,贵阳550025 [2]贵州医科大学附属医院,贵阳550025 [3]贵州大学计算机科学与技术学院,贵阳550025

出　　处：《计算机与数字工程》2025年第3期776-780,810,共6页Computer & Digital Engineering

基　　金：贵州省高层次人才项目(编号:黔科合平台人才-GCC[2022]007-1)资助。

摘　　要：心音准确识别在临床上具有重要应用价值。为提高心音识别准确率,提出了一种基于小波散射变换(WST)和双向长短期记忆网络(BiLSTM)相结合的正异常心音信号分类研究的模型。首先利用小波散射网络自动提取信号特征,将得到的不同维数的特征矩阵输入BiLSTM网络中进行分类处理,为进一步提高准确率,在特征提取前辅以数据增强及自适应降噪处理,并使用主成分分析法对特征参数进行分析并降维,最后将分类特征输入BiLSTM网络中进行训练与验证。相关实验结果表明:该方法的分类准确率达到了89.40%,对比其他模型,具有较优的分类效果。It's important to achieve accurate recognition of heart sounds in clinical practice.To improve the accuracy of heart sound recognition,a model based on the combination of wavelet scattering transform(WST)and bi-directional long and short-term memory network(BiLSTM)is proposed for the classification of positive and abnormal heart sound signals.Firstly,the wavelet scattering network is used to automatically extract the signal features,and the obtained feature matrices with different dimensions are input into the BiLSTM network for classification process.To further improve the accuracy,the data enhancement and adaptive noise reduction processing are supplemented before the feature extraction,and then the principal component analysis is used to analyze the feature parameters and reduce the dimensions of them,and finally the classification features are input into the BiLSTM network for training and validation.The relevant experimental results show that,this method reaches a classification accuracy to 89.40%,and it has a better classification effect compared with other models.

关 键 词：心音 信号分类 小波散射变换 双向长短期记忆网络 

分 类 号：TP391[自动化与计算机技术—计算机应用技术] R318[自动化与计算机技术—计算机科学与技术]