Research On Calibration Transfer And Wavelength Selection Method In Near-infrared(NIR) Spectroscopy Analysis

Because of the advantages of fast speed, high efficiency, non-destruction and easy on-line measurement, nowadays the near-infrared(NIR) spectroscopy analytical technology is widely used in many fields. But in the practical application, the spectra measured on various instruments will have differences due to the aging equipment and the difference of measurement environment. The above situation makes the calibration model established in one instrument cannot be used in the other instrument, while the model reconstruction will cause huge waste of manpower and material resources.Moreover, near-infrared spectral quantitative analysis relies on the multivariate calibration model, and the quality of the model depends on the quality of wavelengths. Due to serious spectra overlapping, information redundancy and unobvious characteristic absorption region, the quality of the model will be poor.This paper researches the above problems and proposes effective solutions. The concrete research contents are as follows:(1) In order to solve the calibration transmission problem in near-infrared(NIR) spectroscopy, a novel model transfer method——SLRDS based on Simple Linear Regression have been presented in this paper. To validate the validity of the proposed method, two real NIR dataset were tested and the direct standardization(DS) method and piecewise direct standardization(PDS) method were involved as a comparison. Our results indicated that SLRDS can eliminate NIR data differences among instruments and enable the user to share PLS calibration model between instruments, at the same time it has higher prediction accuracy, fewer parameters and simpler model than DS and PDS.(2) In near-infrared(NIR) spectroscopy, the serious spectra overlapping, information redundancy and unobvious characteristic absorption region will lead to a poor model. To address this problem, we propose a new method, called MWCS, for wavelength selection, based on moving window strategy and cuckoo search algorithm. The method used cuckoo search algorithm to select wavelength interval to ensure the continuity of selected wavelengths, while a new moving window strategy was applied to avoid inaccurate positioning of information interval because of fixing the interval position and length in interval selection period. To verify the validity of the method,experiments were conducted on corn data set and gasoline data set, and two traditional methods, successive projections algorithm and elimination of uninformative variables, were involved as a comparison. Results show that the MWCS can try to ensure the continuity of the spectral data while selecting the characteristic wavelength, and has a higher accuracy and stability of wavelength selection and a better model prediction performance.