[The research to the optimal calibration model for discrete scanning spectrometers].

For spectrometers which don't operate in consistently scanning mode such as fixed-filter spectrometer, diode arrays spectrometer and so on, how to find an optimal wavelengths combination rapidly and accurately has always been a key problem in NIR so far. The conventional methods to choose the optimal wavelengths for calibration are Forward Stepwise multiple linear Regression (FSR) and Backward Stepwise Regression (BSR), which choose the optimal wavelengths based on the t test result of all the calibration wavelengths. However, maybe the wavelength eliminated is very useful in application and the wavelength chosen may not be so useful as expected, as a result it is compulsory to make more tests to verify the truly applicable wavelength combination for calibration. In this paper the combination of the combination-making algorithm in combinatorics and computer languages which operate facing matrix is used to choose the optimal combination of wavelengths for calibration automatically and efficiently by computer. The most robust calibration equation can be obtained by making analysis of calibration regression that operates to find the minimum root mean standard error of calibration.

[The algorithm of eliminating the similar samples in the process of calibration and prediction].

During the manufacture and debugging of the NIR instruments it is compulsory to make final calibration analysis to verify whether the instruments operate correctly. However, the conventional method of NIR calibration needs to make calibration analysis for the instruments with all the samples at hand. In fact in order to cover the samples that will be encountered in the future the amount of the samples set and the labor that the personnel need to offer are enormous. In this paper a new algorithm is presented which can be used to effectively eliminate the similar samples in the original sample set. By using this algorithm we have chosen 94 optimal samples in the original 178 sample set successfully. After performing calibration experiment we found that the sample set chosen by this algorithm are equally representative to the original sample set and obtained almost the same precision compared to the original sample set when the two sample sets were individually calibrated. This algorithm brings great relief to the labor of the workers, presents the possibility of performing more experiments and greatly improves the efficiency of performing calibration experiment. As a result, the amount of calibration sets and the labor of the personnel are reduced remarkably.