CHENG Xiefeng1,2*，HUANG Jianzhong1
(1 College of Electronic and Optical Engineering, Nanjing University of Posts and Telecommunications, Nanjing 210023, Jiangsu, China;2 National and Local Joint Engineering Laboratory of RF Integration & Micro-Assembly Technology, Nanjing 210023, Jiangsu, China)
Abstract:
In the feature extraction of heart sound signals, most researchers use a method of segmenting heart sounds. However, due to the difference between pathological heart sound and normal heart sound, it is difficult to find a way to accurately segment all kinds of heart sound signals. Therefore, this paper proposes a classification recognition method without segmentation. The signal is filtered and it is extracted the autocorrelation function from the envelope of the heart sound homomorphic feature and saved as an image format. Then, according to the characteristics of the heart sound, the convolutional neural network (CNN) is designed as a classifier. Finally, training, verification, and testing are performed. The experimental results show that the recognition rate is 100% on the verification set and the modify accuracy is 90.21% on the test set.
KeyWords:
heart sound; homomorphic feature envelope; autocorrelation function; convolutional neural network
