The Application Of Artificial Neural Networks On Recognition Of Pulse Signals

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, the investigation of objectivity of the Chinese medicine Pulse- feeling become possible.Artificial neural network (ANN) is a class of engineering system, which imitates the structure and aptitude action of human cerebra by comprehending it's structures and running mechanism. ANN has the strong self-adaptability and self-learning-ability as well as excellent robustness and tolerance ability. An optional nonlinear input-output mapping relationship can be achieved through the network. Concrete mapping relationship materializes at the distributed linking weight values between neurons that build up the ANN. A trained neural network may be used to realize pattern recognitions, signal processing and detection, etc.Considering the characteristic differences between the pulse signals of heroin addicts and healthy persons, we successfully use Self-organizing Feature Map (SOFM) network and Probabilistic Neural network (PNN) to identify heroin addicts from the pulse signals of 15 heroin addicts and 15 healthy persons. Firstly, a two-layer self-organizing network with 38~2~2 is constructed in this paper. The input signals of the network are obtained by clipping the characteristic section of every pulse signal. The network is trained through the Kohonen arithmetic. When the 20 training pulse signals are tested in the trained network, all of the heroin addicts and the healthy persons are correctly identified. But when use the trained network to the 10 testing pulse signals, experiment shows the pulse signals are identified except that healthy persons Z01 and Z10 and heroin addict B13 are misjudged. The exactness ratio of self-organizing network reaches to 90%. When the 30 original pulse signals are trained and tested in the PNN, the exactness ratio reaches to 96.7%.The research shows that the exactness ratio is high, the net structure is simple and the training speed is fast when the PNN analyses the pulse signals. And the research also shows that the training speed of the SOFM network is fast, but the exactness ratio is inferior to that of the PNN.Besides the theories and algorithms of the SOFM network and the PNN are deduced and proved in this paper, and the basic theories coherent to neural network are collected and generalized, which includes the classification of the neural network structure, the algorithms and the training function.