Recognition Of Heart Sound Signal Based On ITD And Improved MFCC

In recent years,all aspects of the human lifestyle are constantly changing.The rate of cardiovascular disease initiation and death in the world has been very high,which poses a serious threat to human health.The sound produced by the mechanical vibration of the heart is called heart sound The early stage of organic heart disease is accompanied by the change of heart sounds and the appearance of heart murmurs,so heart sounds are closely related to cardiovascular diseases.In the past few decades,heart sound signal analysis has been widely studied in clinical applications.As a physiological signal,the heart sound signal is susceptible to various environmental noises and equipment noise during the acquisition process.Therefore,the heart sound signal denoising has always been a hot research topic.In addition,the heart sound signal segmentation process is analyzed in the heart sound signal Plays an important role in this,however,due to the non-stationary nature of heart sound signals and the characteristics of being susceptible to noise,heart sound segmentation is always a problem.It is gratifying that existing literature has shown that accurately segmenting the period of heart sound signals is not heart sound recognition.The necessary conditions for extracting the characteristic parameters that conform to the physiological and pathological essence from the heart sound signal are the key steps of heart sound recognition.Based on this,this article mainly does the following work:For heart noise signal denoising,first use Butterworth band-pass filtering for preliminary denoising,and then according to the characteristics of some of the more difficult to remove noise spectrum and heart sound signal spectrum overlap,this paper proposes an iterative threshold heart sound based on wavelet packet decomposition Signal denoising algorithm,this algorithm uses a combination of wavelet packet decomposition and iterative reconstruction,and adaptively selects the denoising threshold during each iteration,continuously filtering noise through iterations,thereby obtaining pure heart sound signals.For the study of unsegmented heart sound classification,this paper applies Intrinsic Timescale Decomposition(ITD)to heart sound signal analysis to extract the different frequency components contained in the heart sound signal,and the characteristic parameters in these frequency components are improved by using Mel Frequency Cepstrum Cofficient(MFCC).Finally,select a Long Short Term Memory Network(LSTM)classifier suitable for processing timing signals.The final experiment shows that the recognition rate obtained by the extracted feature in thispaper is superior to the recognition rate obtained by other feature parameters.