Research On Feature Extraction And Pattern Classification Methods Of Meridian Electrical Impedance Signal

The meridian electrical signals involve a great deal of human pathological and physiological status information, to understand the complex dynamical behavior is of great significance for the meridian scientific research and clinical diagnostic application. The modern study of meridian and acupoint is a developing and crossing discipline, which is based on the combination of Traditional Chinese Medicine Science and Information Science to obtain the internal laws of human meridian and its associated characteristics with the human physiological state by extracting its electrical impedance characteristics, thus to explore the essence of meridian ultimately. At present, many achievements have been made by the scientists, but the study is still in progress, much work should be done. On the one hand, the meridian electrical signal is the very weak signal which can be described as nonlinear, non-stationary and spectrum-of-time-varying. It is very difficult to be detected and collected effectively and accurately. On the other hand, there has no effective and targeted processing and analysis methods for the studies, especially the lack of useful ways to extract electrical feature and classify patterns on meridian and acupoint, which leads to its conclusion is not ideal for optimal results.Concerning about the problems, the main studies and innovations in this thesis ahead in the following aspects:■Based on bioimpedance technology, an electrical impedance measurement system for human meridian signal was proposed through designing hardware modules, data acquisition modules and incentive source, and the sources of its interference noise were analyzed and solved by a suppression program with software and hardware line. For Solving the distortion problem of the excitation signal, A removal method based on robust matched filtering was simultaneously proposed, and the uncertainty spectrum model was establish by Bashi Wa theorem and its saddle point was saluted by two-Layer genetic optimization algorithm in the method, and Finally its theoretical analysis was given. ■The electrical impedance feature extraction of single meridian signal was investigated. For obtaining its purpose to the dimension reducibility and the maximum utilization and the improvement on accurate recognition, an electrical impedance feature model of single meridian was established based on the AR parameter model, and on which the different order determined criteria and the model solution algorithms were analyzed and compared. Then structural features of the model spectrum were reconstructed and which solved the problem effectively, that is, the AR parameters can only expressed part of system information. In addition, in order to solve the redundant information problem of single-meridian original feature set, the optimization method based on the genetic algorithm for the single meridian electrical impedance signal was proposed. In the method, encoding scheme used binary encoding, fitness function using the within-and-between-class distance criterion, and the individual is generated by selection operator, crossover operator and mutation operator.■The electrical impedance feature extraction of multiple meridian signals was investigated. For improving the data compression rate and reduce feature dimensions and computational complexity, the multi-meridian linear feature extraction algorithm based on PCA was presented, and in which two principal components solution algorithm based on covariance matrix or correlation matrix were compared. Then, the original input vector is mapped to a high dimensional feature space, and the multi-meridian nonlinear algorithm based on KPCA was presented, and in which kernel function was selected and recalculated. Finally, a new comprehensive feature index for the channel based on the PCA method and projection method was established, and in which the ideal Eigenvectors of the original feature set was given by the objective weighting method and projection value of amendments weighted feature matrix was constructed based on the projection of main constituents, Experimental results show that the comprehensive feature index was more effective for pattern classification than single meridian feature index.■The pattern classification on the electrical impedance feature of meridian and acupoint was investigated. For the classification and identification problems of multi-class model, based on Statistical Learning Theory, Empirical Risk Minimization was analyzed and the SVM modeling algorithms for different classification problems were compared, and then a pattern classification method based on LS-SVM for the electrical impedance feature of meridian and acupoint was presented. In the method, grid search method and K cross-validation method were used to solve the problem about approximate optimal parameter of LS-SVM, and various multi-class device methods were discussed. The experimental result showed that the pattern classification based on Directed Acyclic Graph More was effective to improve the recognition rate.■The overall evolution feature of the meridian system was investigated. The systematic characteristics of meridian were analyzed and proposed, and it was pointed out and demonstrated that the meridian system is a primitive information system with relationship existing. Secondly, on this basis, an overall evolution model based on Cellular Automata for the meridian system was presented. In the model, meridian cellular and its space was obtained by the meridian system discredited in time, space and state, and cellular evolution rules in the state transition of single meridian cellular, associated conversion of the channel and historical condition of the meridian were formulated. The simulation results showed that the model can effectively simulate the meridian system with the characteristics of self-loop, self-organization and self-evolution.