Analysis And Recognition Of Heart Sound Signals Based On Wavelet And HHT

Cardiovascular disease has been known as "the first killer" to threaten human health since twenty-first Century, the heart disease incidence rate is increasing higher at current time. Timely diagnosis of heart disease as well as to improve the diagnosis rate is very important. Heart sound is produced by the vibration of heart wall and large vascular wall caused by sudden acceleration or deceleration of blood due to heart valve closure and myocardial contraction. Heart sound signal contains a large number of detail of the physiological and pathological characteristics in human heart, compared with electrocardiogram, phonocardiogram that using electronic instrument to convert cardiac vibrations into current and amplified recording provide more reliable, continuous and dynamic observation to the diagnosis of heart disease,which has higher diagnostic accuracy to certain heart diseases and can diagnose cardiovascular disease effectively at current time.This paper complete the heart sound pretreatment based on the wavelet threshold denoising, using Hilbert Huang transform(HHT) to proceed the frequency domain feature extraction of heart sound signals、segmentation and time domain feature parameter extraction of heart sound signal, and finally through the support vector machine(SVM) to recognize and classify heart sound signal, getting the high recognition rate.In this paper, the main research contents are as follows:(1) On the basis of analyzing the basic principle of wavelet transform, realized the heart sound signal decomposition and reconstruction based on wavelet, compared several kinds of wavelet threshold denoising function, selected the wavelet threshold denoising function whose properties between the soft threshold and hard threshold, completed the heart sound signals pretreatment, simulation experiments show that the method can achieve the better denoising effect.(2) Completed simulation experiment of heart sound signal classic time frequency analysis method based on STFTn Wigner-Ville distribution and HHT, used HHT method with adaptive and local characteristics to analyze nonlinear nonstationary heart sound signal, chose the empirical mode decomposition(EMD) algorithm based on HHT to complete the heart sound signal decomposition, achieved its frequency domain characteristics parameter extraction.(3) Summarized the commonly used automatic segmentation algorithm of heart sound signals, chose the EMD algorithm to extract the heart sound signal envelope, made envelope becomes more smooth through three times spine interpolation,gave the heart sound segmentation strategy, defined the peak position and duration time of the first heart sound (S1) and second sound (S2), completed the time domain characteristic parameters extraction.(4) Based on analysis of support vector machine (SVM)principle, constructed heart sound signal recognition and classification process; completed heart sound signal classification comparison experiment based on the RBF kernel function and Sigmoid kernel function in30cases with normal and abnormal heart sound signal samples; finally achieved the experiment of increasing the training set and test set of samples for classification and recognition of heart sound signals, got high recognition accuracy rate.