Research on automatic heart sound segmentation algorithm based on feature fusion envelope

Aiming at the problem of effective segmentation of heart sounds, a heart sound segmentation algorithm based on multi envelope feature fusion was proposed in this paper, which used interval search combined with threshold judgment to locate Si and S-2. Firstly, the PCG signal was denoised by multi-scale wavelet soft threshold. Then the normalized Shannon energy envelope and the Viola integral envelope were extracted for the denoised signal. The Pearson correlation coefficients between the two and the original signal were calculated respectively. And then the fusion was carried out according to the correlation. Next, the envelope peak points were searched in large interval and between minizones by interval search. After that, the variance of search results was compared and the result with smaller variance was used to compensate for the peak value. Finally, the pseudo peak was removed according to the maximum duration of Si and S2. Through the evaluation of PhysioNet-2016 data set, the Fi score was 0.9015, which proved that this method can effectively supplement the current heart sound segmentation algorithm.