Research Of Pulse Wave Signals Recognition Based On BP Neural Network

Pulse diagnosis is a kind of very important diagnostic techniques in diagnostics of Chinese medicine, which occupies an important position in traditional Chinese medicine (TCM). However, pulse diagnosis has some shortcomings such as great subjectivity, disability of recording objectively, and so on. In the current background of the combining of TCM with modern technology, the pulse objective research has important realistic significance, moreover, it's a new way of thinking on the inheritance and development of TCM. This research is mainly based on the pulse condition identification.First, the theory foundation of pulses identification introduced, we analysed the physical significance of the pulse signal, as well as the working principle of the pulse collecting devices, and then the research extended using pulse signal as object.Second, the research focused on the method and principle of the pulse signal feature extraction, including both aspects of frequency domain and time domain features. Frequency domain analysis method is an important content in the field of signal processing, by using the good time-frequency local character of the wavelet transform, after transforming the pulse signal to frequency signal in the way of wavelet transform technology, it will be pretty convenient to analysis the signal's ingredients in frequency domain. Mainly take multiscale analysis method, decompose the signal in four scales, extract energies in different scales as the characteristic parameters representating different pulses, besides, also extracted the signal's power spectrum character as the identification parameter. Time-domain features is intuitive and easy to understand, with important information in time-domain.This article combined frequency domain and time domain features as the input parameters, which could reflect the characteristics of pulses comprehensively.Finally, we analysed BP network theory and the iteration process based on LM algorithm, LM algorithm converges fast, that could greatly shorten the network convergence time. BP network has the character of nonlinear mapping and adaptive parallel computing, therefore the BP network based on LM algorithm could be used to realize the classification of pulse wave signal. As the research shows, in the conditions of good input vector, it could achieve high identification accuracy.The work done by this article is an attempt to the intelligent identification of TCM pulses, experiments also showed that with every link works well, the identification of pulses can reach precision to a certain extent, and provide certain reference values for clinical, It is a feasible and effective design.