The Research On Segmentation And Classification Of Heart Sound

Heart sound is a kind of biomedical signal,which contains a lot of cardiac activity information,it can reflect the state of heart faithfully,it is the main diagnosis basis of congenital heart disease.The southwest region is one of the high incidence of congenital heart disease,including Yunnan province.so far the main method to select congenital heart disease from healthy person is to rely on experienced doctors auscultation.but it’s inefficiency.A portable device is necessary,which can record and analyze the heart sound signal,give the diagnosis result immediately.this is the purpose of heart sound researchers.In this paper,analysis of normal heart sound and noise characteristics combined with the heart running and heart sound generated mechanism,and a brief introduction to the heart sound signal acquisition equipment.According to the order of heart sound signal processing,introduce the main contents:1.Denoising.In this paper,according to the characteristics of the noise,the wavelet threshold denoising method is used to filter the noise in the heart sound signal.2.Envelope extraction and segmentation.Firstly,the envelope of heart sound signal is extracted by Hilbert transform,and a new method of automatic segmentation is proposed based on the envelope.The results of the experiments were statistically analyzed,and the accuracy of the segmentation results can reach more than 90%.3.Feature extraction.In this paper,a new method of extracting feature parameters is proposed,which is logarithmic energy spectrum(MFSC).MFSC is the Mel frequency cepstral feature(MFCC)to remove the last discrete cosine transform(DCT).The main function of DCT is to solve the correlation processing.In this paper,the recognition method is convolutional neural network,DCT is no need.4.Recognition.CNN is used to identify the extracted parameters,which is a major innovation in this paper.At present,the main methods of heart sound recognition are artificial neural network,support vector machine,clustering and so on.Convolutional neural network has been applied in more and more fields in recent years.The purpose of this paper is to study whether there is an important breakthrough in the recognition of heart sound.More than 300 cases from heart sound database,the correct rate is about 71.2%,Although the recognition rate is low,it is of great reference value,In view of the deep learning use for speech recognition successfully,improve the heart sound recognition rate will certainly achieve good results.