Research On The Human Pulse Identification Based On Multiwavelet Transform

The human pulse diagnosis is an important part of traditional Chinese medicine. Ithas simple, noninvasive and painless characteristics. Therefore, it is concerned andappreciated at home and abroad. However, the traditional pulse identification methoddoes not establish a unified objective standard, which restricts the application anddevelopment of pulse diagnosis. So it is important to extract and analyze the pulsesignal objectively and accurately by modern technology and methods, which hasimportant meaning for realizing the objectification of pulse diagnosis.Multiwavelet is a new development of wavelet theory. As we all know, thesymmetry, orthogonality, short support and high order vanish moments are veryimportant nature of the signal processing. In addition to the Harr wavelet, however, thereal coefficient of a single wavelet can not have all of these properties, which limits theapplication of the wavelet. The multiwavelet can have all of these properties, andtherefore has broad application prospects. The multiwavelet transform is introduced tothe pulse signal processing in this paper, and its basic concepts and theory is described.Features can be extracted by analyzing the pulse signals through the multiwavelettransform method, which can effectively identify differences between heroin addicts andhealthy individuals.In this paper, we analyze pulse signals of15heroin addicts and15healthy personsusing the multiresolution analysis of the multiwavelet transform. After the pulse signalsare decomposed into three levels through the multiwavelet transform method, thefollowing two identification methods are used to distinguish:(1) the signal energy ofeach dimension for each frequency band is calculated, and the feature vector is selected,then we can find out the significant difference between the heroin addicts and thehealthy persons. Z02and Z15are misjudged by this method;(2) the approximateentropy of the decomposition coefficients of each dimension for each frequency band iscalculated, and then the feature vector is selected. By this method, Z15is misjudged.This paper also uses the multiwavelet packet to analyze the pulse signals. Thepulse signal is firstly decomposed into several sub-signals through the multiwaveletpacket transform, and then the coefficient entropy of the sub-band decompositioncoefficients is calculated. Finally, the feature vector is selected. Based on these steps,15heroin addicts and15healthy persons are all correctly identified. At last, the basic concept and theory of the Support Vector Machine (SVM) isdiscussed in this paper. On the basis of the feature extraction of the pulse signals, theSupport Vector Machine network is used to identify the30pulse signals (15heroinaddicts and15healthy peoples), and desired classification results are achieved.