An algorithm of feature extraction for heart sounds based on lifting wavelet transforms

In order to extract the pathological information of heart sounds accurately, an improved method of feature extraction based on lifting wavelet transform analysis was proposed to analyze the first and second heart sounds and recognition different heart sounds purposefully. Firstly, lifting wavelet transform was applied to decrease the noises of different heart sounds by soft threshold method. Secondly, lifting wavelet-time entropy was used to describe the distribution on time domain and extract the entropy. The Shannon energy and improved dual-threshold were then applied to extract the envelope and time of heart sounds. Finally, the best feature elements were analyzed by using SVM, which was used for the classification of sixty different heart sounds. The result showed that these heart sounds are successfully classified with the accuracy of 98%.