Studies On Influential Factors Of Near-infrared Spectroscopy For Chinese Medicinal Materials System

In recent years, near-infrared spectroscopy (NIRS) has been considered as a fast, non-destructive and efficient analysis technique, which is more and more widely used for the rapid quality control of traditional Chinese medicine (TCM). The complete NIRS method includes spectral data collection, model calibration and validation, following model maintenance, updating and transfer. For NIR spectral analysis method, the accuracy of the original data is the assurance to ensure the accuracy and reliability of analytical results. Moreover, NIRS method mainly depends on the delicate differences between the spectral information of samples for further quantitative or qualitative analysis. However, the accuracy of results can be inevitably affected by many influential factors. So the exploration for all kinds of factors and interference is the prerequisite to guarantee the accuracy of NIRS analysis results. Furthermore, NIRS method is a systematic technology developed with the chemometrics; appropriate chemometrics method can be utilized to partially improve the predictability of the NIRS models.This thesis systematically discussed the various influential factors and prevention measures for the near-infrared diffuse reflection spectroscopy and analytical results, including powder particle size, sample temperature, instrument sampling modes and testing conditions, etc. Combined with spectral pretreatment methods, calibration set selecting methods and modeling algorithms, our researches established the anti-interference NIRS models which would be more suitable for TCM system. These work could provide comprehensive and effective guidance for more extensive application of NIRS technology in the rapid analysis field of TCM raw materials. Based on the purposes above, the main research contents and achievements of this study are as follows:(1) Effects of sample particle size on the near-infrared diffuse reflectance spectroscopy for TCM system. Firstly, a variety of TCM materials were used to investigate the effects of particle size on the absorbance of different NIR spectral bands. The results showed that the effects were different on the spectra according to different regions (combination region, first combination-overtone region and second combination-overtone region). In the CR and FCOT regions, spectral intensity was proportional to the particle size; and the influence of particle size was greater as the wavelengths increased. The particle size effects were various according to different types of herbs and different physical structures.Secondly, spectral reproducibility with sample reloading experiment was examined. When sample particle size was smaller than 60 mesh, the spectral reproducibility (RSD<3%) was achieved and well accepted. The smaller the particle size was, the better the reproducibility in all spectral regions would be.Finally, the effects of sample particle size on the quantitative determination of chlorogenic acid (CA) in Lonicera japonica by NIRS method were explored. PLSR quantitative models with six granularities (40 mesh-140 mesh) were established separately for exploring the effects; and both the appropriate sample set partitioning method and data pretreatment techniques were supplemented to optimize calibration models. The 40 mesh model exhibited the worst prediction ability. With the particle size decreasing, the NIRS prediction relative errors also decreased, but not always continued. The mathematical pretreatment methods could partially eliminate the scattering of particle size factors, but could not completely eliminate the interference to the NIRS quantitative models. Furthermore, a global granularity correction model was developed by incorporating particle size variation imformation. It showed better and more accurate predictive performance than each single granularity models.(2) Study of sample temperature effects and correction methods for NIRS quantitative analysis in TCM system. This study investigated the effects of sample temperature variations on NIR diffuse reflection spectra and the prediction models for panax notoginseng saponins (PNS) content in Panax notoginseng samples. NIR spectra of samples acquired at 5 different temperatures (5℃,10℃,20℃,30℃ and 40℃) were very susceptive to sample temperature fluctuations. Not only indicated from the intensity of diffuse reflection, but also in the position of the characteristic absorption wavelengths. Furthermore, sample temperature effects on the predictive performance of constant temperature models were significant, especially for the high temperatures. There was a poor interoperability between 5 different temperature models. Finally, two different temperature correction systems with inclusion of various sample temperature imformation were then established based on PLSR linear method, BP-ANN non-linear method combined with PCA. The predictive performances were further compared and evaluated. Compared with the PLSR temperature correction system, the PCA-BP-ANN correction system did achieve a relatively higher predicted accuracy. The root mean square errors of prediction (RMSEP) for each temperature were reduced significantly. The results indicated that this new multivariate information correction system based on BP-ANN non-linear method had a more comprehensive and accurate prediction ability.(3) NIRS technology is the effective integration of spectral instruments, appropriate chemometrics methods and practicable application models. The spectrometer with various sampling accessories is the important hardware foundation among them. Spectral variances can be reflected between different sampling modes and conditions. Based on that, in this study, the spectra of same TCM materials(Panax notoginseng) were collected by Fourier transform NIR spectrometer with integrating sphere mode and optical fiber mode. The suitable chemometrics methods have been carrying on the spectra of different quality, and predictive results were compared with the statistical test. The results showed that the fitable data processing method can partially improve the prediction accuracy of a weaker performance testing mode, and closed the gap with the high performance mode. In order to further realize generality of the quantitative models established by NIR spectra in different sampling modes, piecewise direct standardization (PDS) method was used under optimal parameters for the implementation of model transfer for Panax notoginseng samples in different sampling modes. At the same time, both combined with spectral processing methods to remove random noises and unrelated interference information. The results showed that PDS method could implement the data model transfer effectively, and improve the precision of the NIRS analysis model.