Research On Heart Sound Classification Method Based On CNN And LSTM Network

At present,cardiovascular disease is the disease with the largest number of patients in my country,and its mortality rate ranks first among the disease types of residents in the country.Heart sound signals contain characteristic information that characterizes cardiac function,which can be used for the prevention and diagnosis of cardiovascular diseases.This paper analyzes the original heart sound signal,uses the time-frequency characteristics of the heart sound signal,and studies the noise reduction and classification of the heart sound.The main work and contributions are as follows.(1)In the aspect of heart sound noise reduction,a heart sound noise reduction method based on CEEMDAN and optimal wavelet is proposed.Aiming at the loss of high-frequency effective information in wavelet denoising,the CEEMDAN adaptive decomposition algorithm is introduced to improve the signal distortion;for the difficulty in selecting wavelet base parameters,the optimal wavelet algorithm is used to adaptively select the optimal wavelet base.First,the original heart sound signal is decomposed by the CEEMDAN algorithm to obtain its various-order IMF components;secondly,the threshold demarcation point of the IMF component is determined by the autocorrelation function method,the noisedominated IMF component is discarded,and the aliased modal component is used.The optimal wavelet method is used to denoise;finally,the denoised IMF components and lowfrequency IMF components are reconstructed to obtain denoised heart sound signals.The noise reduction method in this paper has achieved good noise reduction effect on the three datasets,which verifies the effectiveness of the noise reduction algorithm in this paper.(2)In terms of heart sound classification,a heart sound classification method based on CNN-LSTM is proposed.Aiming at the problem that traditional heart sound classifiers generally need to perform segmentation processing and then extract heart sound features,the neural network-based heart sound classification method is used,which can achieve good classification results without the need for heart sound segmentation.This paper builds three heart sound classification models,CNN,LSTM,and CNN-LSTM based on the deep learning method,and evaluates the classification effects of the three models.Two heart sound datasets are used to verify the classification effects.The noise reduction processing is carried out by using the noise reduction method of CEEMDAN and optimal wavelet proposed in this paper,and the data set B is the original heart sound data set.In this paper,the MFCC algorithm is used to extract the features of the heart sound signal,and the feature vector is sent to the three classifiers for training and testing their performance.The experimental results show that the three classifiers designed in this paper have achieved more than 90% accuracy on data set A.Compared with other methods that also use data set A,the results show that the classification method in this paper has higher accuracy,which can prove that this paper The validity of the classification method;data set B is a small sample data set,which is not universal,but it can verify the effect of the classification model in this paper.The model is valid.