关键词: 信息处理技术, 脉搏波信号分类, 卷积神经网络, 时频域特征混叠, 小波变换

Abstract:

A pulse wave classification algorithm of low complexity based on aliasing of time-frequency domain feature was proposed to solve the problems of low recognition rate and complex implementation in current research of pulse wave signal recognition. First, the time domain features of the pulse wave signal are extracted based on the Convolutional Neural Network(CNN), including the single-period feature characterizing the features of the signal segments in the period and the multi-period features characterizing the relationship between the cycles. Then the features in the frequency domain are expressed by the Mel cepstrum coefficients based on the wavelet transform. Finally, based on the aliasing and redundancy elimination of the time-frequency domain features using the neural network full connection layer, the pulse wave signal classification is realized by the Softmax classifier. The method can achieve feature extraction through low computational cost, due to the weight sharing and dimensionality reduction of CNN. In the simulation experiment based on Python platform, the recognition accuracy of the proposed method can reach 93%, which is much higher than the accuracy of traditional recognition algorithms based on time domain features or frequency domain features.

Key words: information processing technology, classification of pulse wave signal, convolutional neural network, aliasing of time-frequency domain features, wavelet transform