基于提升小波变换的心音信号特征提取方法的研究

为了准确提取心音信号的病理性信息，提出了一种基于提升小波变换的改进的特征提取方法针对性地分析第一心音(S1)和第二心音（S2）及其时限并对不同心音信号进行分类。首先利用提升小波软阈值降噪法对不同心音信号作去噪预处理；然后利用提升小波时间熵法检测心音信号在不同时刻的分布情况，并提取其熵值；通过香农能量优化双阈值法提取心音包络信号及S1、S2时限；最后改进选取心率、S1和S2时限、心动周期、包络面积，熵值六个特征参数，并利用支持向量机算法（SVM）对不同心音信号进行分类。分析和仿真结果表明该算法对正常和心脏病患者的心音准确分类率达到98%，表明该算法能有效识别不同心音信号。

An algorithm of feature extraction for heart sounds based on lifting Wavelet transforms

In order to extract the pathological information of heart sounds accurately, an improved method of feature extraction based on lifting wavelet transform analysis was proposed to analyze the first and second heart sounds and recognition different heart sounds purposefully. Firstly, lifting wavelet transform was applied to decrease noises of different heart sounds by soft threshold method. Secondly, lifting wavelet-time entropy was used to describe the distribution on time domain and extract the entropy; the Shannon energy and improved dual-threshold were then applied to extract the envelope and time of heart sounds. Finally, the best feature elements were analyzed by using SVM, which was used for the classification of sixty different heart sounds. It was observed these heart sounds were successfully classified as the accuracy was 98%.