Research On Human Meridian Potential Signal Processing And Physical Status Classification Method

There are a few similarities between some characteristics of human bioelectricity and some part of Traditional Chinese Medicine (TCM) theory so the study of TCM theory with modern science and technology was introduced to the researchers. The study of human skin electrical impedance characteristics and human potential characteristics are two very important research directions in this area. This thesis provide a method to distinguish and classify the different physiological state of stroke patients and healthy persons by meridian potential signal processing and machine learning theory, which belongs to the second direction.The main contents of the thesis are as follows:1) Collection of human potential signal. According to TCM theory, the properties of "twelve original acupuncture points" change to reflect changes of the state of human body's internal organs. So this thesis used Medlab biological information acquisition system to collect potential signal of 6 original acupuncture points out of the total 12 ones, which are on the hands of body and can be located according to the Chinese National Standards GB 12346-90, for processing and research.2) Wavelet decomposition, de-noising and reconstruction of the collected signal. Wavelet analysis can provide a time-frequency analysis in various scales and is suitable for analysis to the human potential signal, which is distributed mainly in the low frequency range. After removing the part of noise-controlled coefficients from all the wavelet coefficients which were achieved by Wavelet decomposition, the left coefficients were directly used to construct the class I feature vector or to reconstruct the de-noised signal, which is then used to construct the class II feature vector with AR model power spectrum estimation basing on the Burg algorithms.3) Classification of the two types of feature vector, using machine learning methods. All the feature vectors were divided to training set and testing set with Cross-validation method for classification and prediction, using BP neural network, wavelet neural network and support vector machine with different kernel functions. The final results of prediction were compared and analyzed and it showed that the classification accuracy rate of support vector machines for the class II feature vector is very high, indicating the rationality and feasibility of this method for distinguishing and classification of different physiological states for stroke patients and healthy persons.4) At last, the main contents of all sections of the thesis were reviewed. Some problems of the research process were discussed, as well as the future research directions.