The Research Of Deep Learning Algorithm And The Construction And Optimization Of The Heart Sound Deep Recognition System

Heart sounds can reflect whether the heart is beating normally or not, they are closely related to the health of the heart or the human’s body. As a kind of valuable biological signal, now heart sounds widely attract the attention of researchers both at home and abroad. Deep learning algorithm shows excellent characteristics when dealing with big data in complex natural environment, now deep learning algorithm has become the mainstream of the image and voice recognition algorithm.In traditional heart sound signal recognition, the heart sound signals are used in the laboratory,generally speaking, the signals usually need complicated pretreatment. In order to improve the practical applications of heart sound recognition algorithm in big data environment, this paper combined deep learning and heart sound recognition technology, put forward a kind of the heart sound deep recognition system.First of all, in order to solve the problems that the deep learning network structure is difficult to select, this paper delved into the characteristics of the deep learning network’s structure, and proposed a preferred method to help choose the deep learning network structure. Through experimental verification, this method had good effect in a variety of databases, method had a certain universality. Then this paper used the preferred method to build a kind of heart sound deep learning network, the data deal by the heart sound deep learning network can be divided correctly is38.7% higher than the original data, use the selected deep learning network as the core design to build the heart sound deep trust network, compared to other structure networks, the heart sound deep trust network had lower error recognition rate, average error recognition rate is about 10%.To further improve the system recognition rate, improve the heart sound deep trust network features, customize the most suitable classifier for the deep heart sound recognition, the BP neural network, KNN classifier, and other characteristics of the classifiers were analyzed. Finally this paper combined the BP neural network and KNN nearest neighbor classifier, designed a optimized BP neural network which improved BP neural network with poor memory, easy to training mistakenly. Verified by experiment, using the improved classifier to build the heart sound deep recognition system, the system’s error recognition rate fell by 1%, and when the training and testing data library overlap, it will achieve better effect.Finally, although the heart sound deep recognition system had reached the requirements of practical application in the big data, but the recognition rate is still low compared with thetraditional method. So this paper referred into heart sound identification methods in the past,embodied the traditional features of heart sounds into the heart sound deep recognition system, after simply extract, optimizate the original heart sounds energy system to fusing the energy feature deep heart sound identification system. After verified, this system had been a significant reduction in recognition, only about 3%, as traditional recognition method can achieve.