The Application Of Artificial Neural Network In The Multivariate Calibration And Chinese Traditional Medicine Pattern Recognition

There are five parts in this paper. The application of multivariate calibration and pattern recognition of Chinese traditional medicine with artificial neural networks and its development are reviewed in recent years. At the same time, the multivariate calibration methods are applied in the analysis of some complex multi-component systems which can not be analyzed by some routine means qualitatively because physical separation of these systems can not completely accomplished and overlapped peaks of multi-component are formed. Moreover, the pattern recognition method is applied successfully to classify the producing area of Pueraria lobata.In the first chapter, the development and applications of multivariate calibration and pattern recognition of Chinese traditional medicine with artificial neural networks are summarized.In the second chapter, the principles of some chemometric algorithms used in this paper are introduced in detail.In the third chapter, generalized regression neural network (GRNN) and radial basis function network (RBF-ANN) based on linear principal component analysis (LPCA) input selection are applied for quantification of quercetin and rutin in flow injection chemiluminescence. The prediction performance of the calibration models constructed on the basis of two ANN was compared. It can be found that the performances of two ANN are improved after choosing input variables. Moreover, the satisfactory quantitative results are obtained with the use of RBF-ANN calibration model combined with LPCA input selection. In this work, it shows that RBF-ANN based on linear principle component analysis can solve the problems of simultaneous determination of multi-components in flow injection chemiluminescence.In the forth chapter, the nonlinear principal component analysis (NLPCA) is used for the pre-procedure of input variable. The calculated results show that the RBF-ANN based on NLPCA could be applied to determine simultaneously the multi-component in chemiluminescence. In this paper, the satisfactory quantitative results of hydrochlorothiazide and captopril are obtained with the use of RBF-ANN calibration model combined with NLPCA input selectionIn the fifth chapter, artificial neural network is applied to the quantitative analysis of multi-component overlapped peaks in fluorescence. The results indicate that RBF-ANN which uses NLPCA to choose input variables can predict the concentration of each component of overlapped peaks exactly. And it provides a good routine method for the analysis of this kind of sample.In the sixth chapter, self-organizing feature map neural network recognize successfully the producing area of Pueraria lobata. The methods resolved the re-calculation problem of principle component analysis and nonlinear map beginning a new sample. And it is a simple, convenient and exact method for the pattern recognition.