基于CNN的心音特征融合分类方法

Classification Method of Heart Sound Feature Fusion Based on CNN

针对TFF1dCNN方法利用一维CNN分别对各心音片段的4个频带信号提取特征,可能无法充分提取各频带信号间相关信息的问题,提出TFF2dCNN方法。先将4个频带信号融合成二维信号;再由二维CNN进行特征提取和分类。实验结果表明,该方法提升了分类正确率。此外,还分析了心音样本的分类正确率与其包含的心动周期数的关系。

The TFF1dCNN method uses one-dimensional CNN to extract the features of four frequency bands signal of each heart sound segment,and there is a problem that the relevant information between each frequency band may not be fully extracted.This paper proposes the TFF2dCNN method.The four frequency bands signal are firstly combined into a two-dimensional signal,and then the features extraction and classification are carried out by two-dimensional CNN.Experimental results show that the proposed method improves the classification accuracy.In addition,this paper also analyzes the relationship between the classification accuracy of heart sound and the number of cardiac cycles it contains.