改进的高斯混合模型在心音信号分类识别中应用

为提高心音信号特征提取的准确性及分类识别的高效性,将小波包变换的Mel频率倒谱系数与改进的高斯混合模型结合用于心音信号分类识别。在Mel频率倒谱系数提取方法基础上,用小波包变换代替傅里叶变换与Mel滤波器组,获得新特征参数DWPTMFCC;针对传统GMM参数初始化K-means算法缺点,用加权可选择模糊C均值算法进行改进;将提取的特征参数分别输入到改进后GMM进行分类识别。对临床采集的心音数据测试结果表明,该方法能有效提取心音特征,优于传统GMM识别性能。

Application of improved GMM in classification and recognition of heart sound

Chongqing 400044，China)Abstract: To improve the precision of extracting feature and efficiency of classification and recognition from heart sound, the method of Discrete Wavelet Transform Mel Frequency Cepstrum Coefficients (DWPTMFCC) combined with improved Gaussian Mixture Model (GMM) is used for classification and recognition of heart sound. Firstly, the new feature parameter is formed by using wavelet packet transform instead of Fourier transform and Mel filter group on the basis of the extraction method of MFCC; Secondly, to overcome the shortcoming of K-means algorithm which is used in the parameters initialization process of traditional GMM, Weighted Optional Fuzzy C-Means (WOFCM) algorithm is proposed; Finally, the feature parameters are input into improved GMM for recognition. The clinical data experimental diagnosis and test results show that the method not only can effectively extract heart sound feature, but also have better recognition performance compared with traditional GMM.