Heart Sound Signal Feature Extraction Based On Wavelet Technology

Currently,cardiovascular diseases have seriously threaten people’s health.Heart sound signal contains a variety of physiological and pathological information of human body.It is particularly important in clinical diagnosis.At this stage,doctors rely on personal ability and experience,so there are individual differences of diagnosis.In this paper,wavelet de-noising and feature extraction of heart sound signals are proposed.Then the support vector machine algorithm is used to intelligently classify and recognize the heart sound signals of patients.This technique has the advantages of non-trauma,low cost and standardization of judgment,so it has important guiding significance for intelligent diagnosis.The main aspects of this paper are as follows:First,the medical principles of heart sound are discussed.Heart sound pretreatment includes the desample filter and the removal of power line interference.A noise reduction scheme based on wavelet shrinkage technique was proposed in this paper.Firstly,the characteristics of heart sound signal frequency were analyzed and Haar,Daubechies,Symlets and Coiflets orthogonal wavelets were studied for contrast in accordance with the principle of frequency band similarity matching.Based on the statistical results,Coif5 wavelet was chosen for the decomposition and reconstruction of heart sound signal.Besides,an adaptive elastic threshold function was designed for wavelet shrinkage.The noise reduction effects were compared under various threshold rules.It was found that the new threshold function had perfect noise reduction influences,especially under Heursure rule when the SNR is less than 50 dB.Finally,support vector machine was used to classify pathologic sound.The recognition rate of right ventricular outflow tract stenosis was 95.56% based on energy characteristics.For the diagnosis of pulmonary artery stenosis and mitral regurgitation,the recognition rate was 90.67% with time domain and energy characteristics.Faro ’s tetralogy was identified by frequency domain and energy,and the recognition rate reached 91.67%.In conclusion,the results show that the wavelet technique can effectively remove the heart sound noise without distorting,while the relative right ventricular outflow tract stenosis,Faro tetralogy and other heart diseases can effectively identify and classify by time domain,frequency domain,energy domain.And provide important reference for clinical intelligence diagnosis research based on heart sound signal.