| The near infrared spectroscopy (NIRS) quantitative analysis methods are a green qualitativeand quantitative analysis method which developed quickly in rescent years. The NIRS analysistechnology is shown the basial and diploid frequency of organic media’s chemical bond such asC-H, N-H, O-H and C-O and so on, their absorbances are weaker and much more complexthan the other kinds of spectra. A powerful calibration method is needed to extract the effectivemessages of the near infrared spectra. Chemometrics is a popular calibration method whichdevelops soon with the computer technology. It plays important role in parsing spectra. It has beenapplied to agriculture, chemical industry, pharmaceuticals and so on. As NIR analysis technologyis a nondestructive, soonly, low cost, simultaneously analysis muti-component, at-line or on-linedetermination, it is greatly attracted by researchers. However, NIR quantitative analysis method isan indirect method. The precondition of applying this technology is developing a stable andcredible calibration models. The establishing of the calibration model is an important process andthis process is complex. In this research, the development and optimization of the quantitativeanalysis model for quantitative analysis of the anise Oil content in Illicium verum Hook. f. wasinvestigated.In this paper, the recent developments of Illicium verum Hook. f., anise oil, near infraredspectroscopy and chemometrics has been reviewed in the first chapter. The radial basis functionneural network was applied in modeling the relationship between NIR and the contents of aniseoil in Illicium verum Hook. f.. in the second chapter. The optimization of the calibration of modelhas beenfurther studied for evaluation the stability and predictive capability of the calibrationmodel. Firstly, the principal components analysis (PCA) method was applied to analysis themessages of the NIR of the samples. And then, the representational samples were selected ascalibration set samples and prediction set samples depending on the scatter plot of the scores ofthe first principal component (PC1) and the second principal component (PC2) which wasobtained by PCA method and the distribution of the contents of the anise oil in the samples.Secondly, all of the samples were separated into two sample sets according to their anise oilcontents and the messages of the spectra. Savitzky-Golay smoothing method, first order derivative standard normalize transfer (SNV) and fast fourier transform (FFT) were used for preprocessingthe spectra respectively. The original spectra and the preprocessed spectra were used for modelingrespectively. Moving window radial basis function neural network (MWRBFNN) was employedto screen out the charateristical wavelengths variables of the original and preprocessed spectra.Each model which was modeled by by selecting the charateristical wavelength variables has beenoptimized by selecting the most suitable number of hidden nodes depending on the degree ofapproachment (Da). During this optimization process, the effect of size of moving window wasinvestigated. Finally, the suitable spread constant and the number of hidden nodes were selecteddepending on Da. And then, the optimum model had been established. The anise oil contents inIllicium verum Hook. f. has been predicted by this model. RMSEC was0.000, and RMSEP was0.411. Rc and Rp was1.0000and0.9837respectively. These results indicated that this method wasexact. It will be recommented to be a popular analysis method in the pharmacy. The last chapter inthis paper comprehensively summarized this research. |
PDF Full Text Request