Study Of Tonic Traditional Chinese Medicine Classification Methods Based On Near Infrared Spectroscopy

Due to its unique advantages, near infrared spectroscopy has been widely used in traditional Chinese medicine analysis fields.However, in terms of traditional Chinese medicine classification and identification, the insufficiency of sample type, limited range of data reliability and other more serious problem in current researches have given the necessity to establish corresponding database and systematic identification methods for some common traditional Chinese medicinal materials. Moreover, because of the ingredients complexity of traditional Chinese medicine, it is difficult for conventional methods to extract out the information contained in near-infrared spectrum, not mention to give a specific spectral classification basis.So,this paper deeply studied near infrared spectroscopy qualitative analysis technique in order to provide a theoretical basis and technical support for the system classification of traditional Chinese medicine. The main study were as follows:1.Study relevant principles and applications of the near infrared spectroscopy analysis technology in the field of traditional Chinese medicine in depth, focusing on sample selection, spectral pretreatment, wavelength optimization, qualitative analysis and other aspects.2.The near infrared spectroscopy classification model of KNN, BP artificial neural network, RBF artificial neural networks and SVM of18kind tonic Chinese medicine were established. Results showed that k-nearest neighbor method, BP neural networks and support vector machines were more effective nonlinear classification methods, among which support vector machine method was most effective. In least squares support vector machines, best model was obtained when minimum output encoding as multi-class encoding, the number of principal components was6, penalty coefficient y and kernel parameter a2were2.1007and1.9604. Calibration set and prediction set discriminant rates reached95.37%and95.74%respectively, showing outstanding stability and generalization ability.3.Relevant theories and methods of applying latent semantic analysis (LSA) into near infrared spectroscopy analysis of tonic Chinese medicine were proposed. Four basic elements in latent semantic analysis text model, words, text, question and vocabulary-text matrix,were given new related definitions in near infrared spectroscopy analysis model. Meanwhile, calculation process and formula were deduced in details. Calculation criterion function of vocabulary-text matrix and absorbance intensity threshold which affected model dramatically also were discussed emphatically.4.Feasibility and classification effectiveness of the latent semantic analysis method applied to the near-infrared spectral analysis were testified through18kinds of tonic Chinese medicine. LSA classification model obtained from adjust modeling parameters λj and βj, selecting space layers k did not only required very few pre-treatments, but also had better classification results than that of KNN, BP artificial neural network and LS-SVM methods with tis recognition rate of calibration set and prediction set96.85%and97.59%.5.The near-infrared spectroscopy classification basis of18kinds of tonic Chinese medicine were obtained via statistical analysis based on latent semantic analysis methods.