The Detection Of Pulse Signals Based-on RBF Neural Network

The diagnostic methods and particularly curative effect of traditional Chinese medicine have been playing an important role of national health care. Pulse-feeling is one of the primary diagnostic methods in traditional Chinese medicine. Along with the development of sensors and computer technology, people hope to apply modern signal processing technology to human pulse diagnosis, in order to carrying on an investigation in the objectivity of the Chinese medicine Pulse-feeling, which is the basis of this paper.Radial basis function neural network (RBFNN) is a three-layer forward network whose output nodes form a linear combination of the basis functions computed by the hidden layer nodes. The basis functions in the hidden layer produce a response to input stimulus through transfering the input signals in lower dimensional space into higher space. The transference makes the input signals be linearly separated easy. RBF neural networks have been extensively used in such diverse fields as time series analysis, image processing, pattern recognition and so on, owing to their features of simple architectures and brief training requirements.Considering the characteristic differences between the pulse signals of heroin addicts and healthy persons, we successfully use RBF network and RBF network optimized by genetic algorithm to identify heroin addicts from the pulse signals of 15 heroin addicts and 15 healthy persons. Firstly, a RBF network with 40～16～1 is constructed in this paper. The input signals of the network are obtained by clipping the characteristic section of every pulse signal. The basis functions in the hidden layer are identified by K-means clustering algorithm and weights of the output layer are trained by pseudoinverse method and least mean square algorithm. The trained network arrives at a 100% identification rate for 20 training samples and a 90% identification rate for 10 test samples. The identification rate for the all of 30 pulse signals arrives at so high as 96.7%. Only one healthy person Z10 is misjudged in all the 30 pulse signals. The number of the hidden nodes and the basis functions in the hidden layer are identified by genetic algorithm. Objective function of the populations in the genetic algorithm using Akaike's information criterion is combined objective values of training set with objective values of test set. After iteration, the best architecture of the RBF network with 40～19～1 is found. This network can correctly identified the all of 30 pulse...