Recognition And Classification Of Heart Sound Signals Based On Envelope Extraction

Heart sound signal is one of the most important physiological signals of human body,and all kinds of information related to cardiac disease are often reflected in the heart sound signal.The change of heart sound signal and the appearance of murmur are often the earliest signs of organic cardiac disease.Furthermore,heart sound signal detection is convenient,noninvasive and underspend,comparing with the electrocardiogram,ultrasound cardiogram,etc.Therefore,the quantitative analysis of heart sound signal is of great significance for basic research and clinical diagnosis.In this dissertation,the methods of envelope extraction and segmentation of heart sound signal are studied.The main research contents are as follows:Firstly,this dissertation preprocessed the heart sound signal.On the one hand,in order to reduce the amount of data processing for subsequent analysis,the heart sound signal is re-sampling.On the other hand,Butterworth band-pass filter and wavelet threshold de-noising method are used to remove the noise of different frequencies from heart sound signal.In partic-ular,since the heart sound signal is approximately periodic,a new method is proposed to obtain the signal of heart sound with uniform strength and length,prepare for subsequent segmentation and recognition.Secondly,a new model of envelope extraction is proposed,which has obvious advantages over the existing envelope extraction method.For mono-component signal,the upper and lower envelopes obtained by the model can guarantee their symmetry.The intrinsic modal function of empirical mode decomposition is an empirical model of a mono-component signal,so the new envelope model is combined with the empirical mode decomposition to extract the envelope of the intrinsic mode component that can represent the heart sound signal.The excellent features of the new method can be seen from the subsequent segmentation result of the heart sound signal.Thirdly,in terms of feature extraction,this dissertation proposes a new heart sound signal segmentation method based on the envelope model proposed in Chapter 4,which includes two steps of cardiac cycle estimation and heart sound segmentation.The cardiac cycle estimation of the heart sound can guarantee the timely correcting when there is a large deviation in the subsequent segmentation process,and improve the accuracy of segmentation.Then extract the time domain features by segmentation.In addition,this dissertation also extracts the frequency domain features and wavelet coefficient energy features of heart sound signal.Experiments show that a good classification recognition rate can be achieved based on this joint feature.