摘要： 针对信号识别率高低由识别模型及特征参数决定的特点，提出融合K均值聚类的多观察序列的Baum-Welch参数重估算法，用于训练隐马尔科夫模型(HMM)，通过主分量分析(PCA)对梅尔频率倒谱系数进行变换，并设计与实现一套基于PCA和HMM的心音自动识别系统。实验结果表明，该系统对6类常见心音的平均识别率达到83.3%，性能优于其他心音识别系统。

关键词: 梅尔频率倒谱系数, 主分量分析, 隐马尔科夫模型, K均值聚类, Baum-Welch算法, 心音识别

Abstract: According to the signal recognition rate decided by recognition model and the characteristic parameters, this paper puts forward the fusion K-means clustering of the observed sequence Baum-Welch parameters estimation algorithm to train Hidden Markov Model(HMM), the Principal Component Analysis(PCA) is adopted to transform Mel Frequency Cepstrum Coefficient(MFCC) features. A heart sounds signal automatic diagnosis system is designed based on PCA and HMM. Experimental results show that the average recognition rate of 6 common classe’s heart sounds reaches 83.3%, the performance is better than other heart sound recognition systems.

Key words: Mel Frequency Cepstrum Coefficient(MFCC), Principal Component Analysis(PCA), Hidden Markov Model(HMM), K-means clustering, Baum-Welch algorithm, heart sounds recognition

中图分类号: 

• TP391