近红外水分稳健分析模型研究

样品水分含量差异对近红外光谱分析模型的稳健性影响最为严重。文章以全籽粒小麦蛋白质含量为研究对象，分析了光谱预处理、有效波数区间的选取和全局校正模型应用对建立近红外水分稳健分析模型的可行性。结果表明，仅通过光谱预处理方法不能减少样品水分差异对预测结果的影响;选择有效波数区间和建立全局校正模型对消除水分的影响均有效，建立全局校正模型的效果最佳。并从理论上初步分析了各种方法的作用机理。

Study on the Robust NIR Calibration Models for Moisture

The differences in sample moisture affect the robustness of NIR model obviously. In the present paper, three approaches, including preprocessing spectra, selecting wavelength, and setting up global calibration, were investigated to analyze the feasibility of setting up robust calibrations based on the protein content of wheat with different moistures. It has been found that with only spectral pretreatment method it fails to obtain satisfactory results, which can not remove the effects caused by moisture difference. Both selecting wavelengths and developing global calibration model proved to be good approaches for developing robust NIR calibration, yet developing global calibration is better. The mechanisms of the three different methods were also analyzed theoretically.