Study On Multivariate Calibrations Of Near-infrared Spectroscopy For Traditional Chinese Medicine

Calibration models are needed for near infrared spectroscopy(NIR) because the contents of components in sample cannot be measured directly by NIR. Traditional Chinese Medicine(TCM) is a complex system, and for TCM analysis by NIR, it is so important to improve performance of calibration models that methods for qualitative and quantitative analysis, spectra preprocessing, wavelength selection are being developed. The main jobs of this thesis are the following.Firstly, a new wavelength selection method based on the relevant information between chemical component and the NIR spectra is developed .The NIR spectra of Traditional Chinese Medicine are very complex. A mass of calculation will be necessary if all wavelengths of NIR spectra are employed and the precision of calibration model will probably reduce because of the irrelevant information. The proposed method for wavelength selection is successfully used for analyzing Panax notoginseng and Coptis chinensis alkaloids with macroreticular resin by NIRS analysis. The calculated results indicate that the method is a practical preprocessing procedure for near infrared spectra analysis.Secondly, a new calibration method based on BP neural network and feature extraction is introduced to model the quantitative correlations between the NIRS and the contents of Glycine, Arginine and total amino acid in Cordyceps sinensis. By comparing several preprocessing procedures and different wavelength ranges, the optimal models are obtained. The results show this method is effective and it can serve as an excellent nonlinear calibration tool for fast measurement of content of chemical components in TCM by NIR.