Heart Sound Signal Recognition Of Valvular Disease Based On Spectral Features

The heart sound is a biological signal generated by the various organs around the heart during mechanical movement.It contains a large amount of physiological and pathological information in the human body,which can reflect the health of the heart system to a certain extent.Therefore,how to extract effective information from heart sound signals to distinguish different types of heart diseases has become an important issue.Based on this,this paper firstly transforms the pre-treated valvular heart sound signal into different kinds of heart sounds,and then analyzes the heart sound signals of several valvular diseases by extracting the feature quantity of the heart sound spectrum.Finally,the support vector is adopted.The machine acts as a heart sound signal classifier to classify and recognize different types of hearts.The work of the paper mainly includes the following parts.(1)Preprocessing of heart sound signals:In the process of collecting heart sound signals from patients with valvular heart disease,some external noises are inevitably included,which leads to the insignificant characteristics of useful heart sound information.In order to better extract the original features of heart sound signals of different valvular diseases,wavelet denoising method is used to denoise the heart sound signals of different valvular diseases.In the experiment,the original characteristics of the normal heart sound signal and the heart sound signal of different valvular diseases are compared firstly,in order to preserve the effective information of the original heart sound signal as much as possible,and achieve a relatively good denoising effect.This paper uses the energy variance and maximum at each level.Coif5 wavelet is used to process the heart sound signal of valvular disease,and the decomposition layer is selected by comparing the effect of Coif5 wavelet decomposition layer.In the process of noise reduction,the innovation of this paper is to combine the characteristics of soft and hard thresholds to construct a new threshold to compare with the wavelet coefficients.The new threshold overcomes the hard threshold denoising to generate jump points in the denoising process.The problem overcomes the problem of soft threshold denoising bias.Experiments have shown that this method can denoise the heart sound signal of valvular disease to a certain extent.(2)Acquisition of heart sound spectrum:The acquired heart sound signal belongs to the one-dimensional signal.The analysis of the heart sound signal from the one-dimensional angle can obviously observe the difference in amplitude,frequency,etc.of several heart sound signals,but is unique to some heart sound signals.The details are not easily observed.However,if a one-dimensional heart sound signal is converted into a two-dimensional heart sound spectrum for analysis,information that is not easily observed in one dimension can be expressed on the image,so the heart sound signal is analyzed from the perspective of the two-dimensional image.It will provide some ideas and ideas for feature extraction and classification recognition of heart sound signals.As a non-stationary signal,the one-dimensional heart sound signal needs to be windowed in the process of transforming into a two-dimensional heart sound spectrum.In this paper,when selecting the window function,by comparing the advantages and disadvantages of several window functions,the Hanming window with the characteristics of main lobe width and low side lobe height is selected to window the heart sound signal.After the processing,the heart sound signal is processed.The short-time Fourier transform is converted into a two-dimensional image,and the gradation spectrum of the heart sound signal is obtained.(3)Acquisition of heart sound spectrum feature vector:Based on the close correlation between the heart sound signal type of valvular heart disease and the texture distribution of the heart sound spectrum.A feature extraction algorithm based on improved local binary pattern(LBP)is proposed to extract the heart sound signals of several different valvular diseases.The experiment found that the traditional local binary mode algorithm is used to extract the features of valvular heart sound signals,and its recognition effect has no advantage over other feature extraction algorithms.Therefore,this paper introduces a complete local binary pattern which can also extract image features,and improves the complete local binary pattern,and then combines it with the local binary pattern to extract the feature vector of the heart sound spectrum.Experiments show that the feature vector extraction method proposed in the paper can be used to characterize the features of the heart sound spectrum.(4)Classification of valvular heart sound signals:In this paper,when the proposed algorithm is validated,the normal heart sound signal and several valvular heart sound signals are selected as training samples,and the support vector machine is used as the classifier to classify and recognize several heart sound signals.In order to select the appropriate heart sound signal classifier parameters,four kinds of kernel functions,such as linear kernel function,radial basis kernel function,polynomial kernel function and S-type kernel function,are selected for training.Finally,the polynomial kernel with the highest recognition rate is selected by comparison effect.The function acts as an SVM kernel function.Experiments show that this algorithm has a better classification effect on heart sound signal of valvular disease,and it has a significantly higher recognition rate than other feature extraction algorithms,which provides a new idea for the early treatment of cardiovascular disease.