| Near-infrared (NIR) spectral analysis technology has been successfully applied to realize the rapid and non-destructive analysis in many fields. But a thorny problem associated with the calibration model transfer has not been well solved so far. The calibration model developed on one instrument can't be applied to others directly in many cases. The problem that the multivariate calibration models can't be applied to different instruments and can't adapt to the aging equipment has become the main technical bottleneck for the large-scale application of NIR analysis. Therefore, in order to solve this problem, in this dissertation, the main contents are as follows:(1) In the study of the multivariate calibration model transfer of NIR, special attention was paid to the selection of transfer samples. To implement calibration model transfer between the master and slave instruments, an improved piecewise direct standardization called as Rank-KS-PDS is proposed in this paper. In this new proposed method, an optimal selection method (Rank-KS) is applied to select the transfer samples. The process of selection is based on the distribution feature of spectral space and property space simultaneously. So the samples selected by Rank-KS have a higher representativeness and wider coverage. In this work, the effect of the proposed method had been verified by the prediction of alkaloid and glycosides content in tobacco. With the aim of carrying out a model for routine analysis, the partial least squares (PLS) models of alkaloid and glycosides were constructed. For comparison, during the process of calibration model transfer, the transfer samples were all selected by random selection (RS), KS and Rank-KS methods, respectively. Then the PDS method was used to implement the calibration model transfer between different instruments. Compared with the existing methods, the Rank-KS-PDS method can reach the best results and the requirement of transfer samples is at least.(2) Most of the existing method are aimed at realizing the calibration model transfer between the instruments of the same type, and the resolution of the spectra is also the same. However, no in-depth study of the calibration model transfer between the instruments of different types, and between the spectra of different resolutions measured on the same instrument. For this point, a new calibration model transfer method was proposed, called as SP-SD1st-PDS. In this new proposed method, a fitting slave spectrum is constructed by cubic spline interpolation without destroying the information of the original slave spectrum. Then Savitaky-Golay with first order derivative is applied to remove the baseline drift of the spectra. Finally, the PDS method is applied to remove most of the differences between the master and slave spectra. In this work, the new method was applied to predict the content of vinyl acetate (VAC) in ethylene vinyl acetate (EVA).In this experiment, the process of the reach can be departed into two parts: the first part is the model transfer between the instruments of different types. In this part, the master instrument is the SupNIR-2700 spectrometer, and the slave instrument is Nicolet Antaris ?FT-NIR spectrometer. The second part is the model transfer between the spectra of different resolutions measured on the same instrument. In this part, the spectra with different resolutions were measured on the Nicolet Antaris ?FT-NIR spectrometer, and the resolutions was set as 4cm-1 and 8cm-1. The results shows that the effects of the new proposed method can improve the accuracy of the model transfer between the spectra measured on the instruments of different types. The RMSEP of the new method is much smaller than other methods. But for the model transfer between the spectra of different resolutions measured on one instrument, the accuracy of the model only be improved slightly after model transfer. The main reason is that the information of the spectra with different resolution measured on the same instrument is almost the same. |
PDF Full Text Request