Research On Array Pulse Signal Fusion And Classification Methods

Pulse diagnosis is one of the most important bases in syndrome differentiation and treatment of Traditional Chinese Medical(TCM).In classical theories of TCM,practitioners put three figures on the wrist of the patient to analysis the health condition through the fluctuations in the radial pulse.With the dev elopment of sensors,the pulse signal has recently been acquired through the sensors.The study of pulse classification has been carried out by the statistic pattern recognition.These techniques have promoted the objective development of pulse condition.However,a great deal of relevant information will be generated.Pulse features have been extracted through different measures and hold different dimensions.A large number of redundant information will make difficulties in pulse diagnosis.In this paper,we mainly study the array pulse signal fusion and classification methods.The features and channels of pulse signal have been optimized.What's more,a new feature fusion method has been put forward to fuse multichannel and multifeature.The optimization of pulse features and channels is the basic of fusion problems.To some extent,dimensionality reduction and feature selection can reduce the difficulty of analyzing pulse features.According to the evaluation criteria of class separability,the insensitive features to the classification of pulse signals are eliminated.In the preliminary selected features,the optimal selected features can be obtained with the algorithm of Relief-F.The similarity and difference between each independent channel has been studied.The pulse classification results of seven strategies of channel fusion are different.Through the comparisons of classification results,the optimal selection method of pulse channels can be proposed.Here,the best optimal selection strategy is the fusion strategy of three channels at the same time.Based on the optimization of features and channels,an improved method of feature fusion has been proposed.This new method is named as Karhunen-Loeve Multiple Generalized Discriminative Canonical Correlation Analysis.On the one hand,the new fusion method breaks the boundary of multiple features and channels.On the other hand,it integrates to the generalized categorization with discriminative information.Furthermore,this method also provides a solution of pulse feature vector sets with high-dimensional space and small size.The pulse fusion and classification method has been built through the improved method.It can realize the multifeature and multichannel fusion with noninvasive operation.The binary classifications between disease and health have been carried out.Compared with the traditional single feature and channel classification,the traditional canonical correction analysis methods,the serial fusion method and the decision fusion method,our proposed method can improve the classification accuracy effectively.