Study On The Calibration Technology Of Near-infrared Spectroscopy Analysis

Calibration is the key part of near-infrared spectroscopy analysis. It includes the technology of spectra preprocessing and regression. In order to make the calibration model precise and robust, We must choose the optimal method of spectra preprocessing and regression for the specific analyte. How to make these choices arealways the hot topics for the researchers.In this paper, diffuse reflectance absorption spectra of fifty tobacco samples werecollected. MSC, SNV, Second order derivative and OSC were used to preprocess the spectra respectively, then PCR and PLSR were used to develop the calibration models of total sugar and nicotine. Spectra analysis, variance analysis, correlation analysis, principal component analysis and regression analysis were performed to compare all kinds of preprocessed spectra and calibration models.It turned out that: The results of all kinds of analysis for MSC and SNV are very similar, which can give an explanation theoretically why the calibration models developed by the two kinds of corrected spectra are almost the same. The model developed by the OSC corrected spectra is much better than the others for the ingredient of low concentration and low absorbance. The four preprocessing methods all can simplify calibration model but can't improve the prediction ability of model dramatically for the ingredient of high concentration and high absorbance, so we may choose MSC or SNV as the preprocessing method because their algorithms are much simpler than the others. PLSR is superior to PCR in almost all the cases except when using OSC corrected spectra to regress. In this case, the performances of PLSR and PCR are becoming more and more similar with the increase of the numbers of OSC correcting factors.