Research On Digital Pulse Information Acquisition And Recognition Technology

Objectification of pulse diagnose is one of the important subjects intraditional Chinese medicine. The researchers have been hoping for realizing thedigitization of pulse information acquisition and analysis in Chinese medicine, so asto rule out the subjectivity and randomness when the doctors pulse-diagnosing.Based on the above cause, the pulse digital acquisition, characteristic analysis andpattern recognition are newly researched and evaluated.First of all, according to the “Three-Bu-Nine-Hou” pulse diagnosetheory in traditional Chinese medicine, the pressure controllablethree-channel pulse acquisition device is designed. The device cansynchronize the pulse signal acquisition in the testee’s three parts called “Cun,Guan, Chi” in three kinds of pressures called “Fu, Zhong, Chen”, real-timelytransmit the pulse signal data to PC through the USB interface. The PC pulseacquisition and analysis software is developed for real-time pulse acquisitionand display. A lot of different group’s pulse signals is acquired by the deviceand validated to meet the design requirements. Secondly, on the basis ofpulse signal preprocessing such as smoothing, removing baseline drifting andnormalization, according to the description “position, rate, shape, force” oftraditional Chinese medicine pulse, the time domain characteristics of pulsesignals is analyzed and the pulse’s physical characteristic such as frequencyand pulse pressure is extracted. In the frequency domain, the spectrum ofpulse signal is resolved through Fourier transform, the energy distribution ofdifferent pulse signal frequency band is analyzed by the wavelet packettransform and the complexity of different pulse is computed using the sampleentropy analysis method. Finally, using the pattern recognition method, thepulse classifier model is established with generalized regression neuralnetwork(GRNN) and probabilistic neural network(PNN) respectively and thefamiliar four different pulse is recognized. At the same time,the experiment results is classified and analyzed relatively by GRNN and PNN with pulsefrequency band energy, pulse frequency and sample entropy features. For thisexperiment, the effect of probabilistic neural network(PNN) is a bit betterthan that of general regression neural network(GRNN).