A Construction Method Of Biorthogonal Wavelet And Research On Heart Sound Wavelet Neural Network

In the basic of analyzing the theory of wavelet construction and heart sound characters, this paper put forward a fast construction method of biorthogonal wavelet and heart sound wavelet neural network. Both application technology were discussed deeply. Firstly, the article presents the specific construction methods, detail procedures and algorithm flow chart of biorthogonal wavelet; Then, a series of wavelet are constructed when the filter length is N and vanishing moment is N/2 and a series of waveforms are shown when the N is equal to 2, 4,6,8,10; Moreover, according to the principles of selecting wavelet in engineering applications, self-construct wavelet that filter length is 10 and vanishing moment is 5 was selected to process heart sound signals, and this wavelet was renamed as heart sound wavelet; Finally, this paper select heart sound wavelet to denoising and feature extraction on multiple sets of heart sound signals, the result shows that compare to db,bior,sym, self-construct heart sound wavelet shows a better comprehensive effect,for example, the reconstruction error rate is lowest, the SNR is highest, the MSE is the minimum, the smoothness is higher and the average separability increased by 50%.In order to make a fusion of heart sound feature extraction and heart sound recognition into a network for the classification of heart sounds. This article puts forward an effective fusion method of wavelet and neural network for constructing heart sound wavelet neural network. In the hidden layer, heart sounds wavelet as the activation function, a new heart sound wavelet neural network will obtained; Finally selects multiple groups of normal heart sounds signal and premature beat heart sound signals as the experimental object, Compared with morlet wavelet neural network and Mexican-hat wavelet neural network, the heart sound wavelet neural network shows obvious priority on algorithm convergence and speed;the accuracy of recognition rate of heart sound also reached 97%.All this above research will have a very important theoretical significance and actual affect for promoting application of wavelet and heart sound research.