烟草水溶性糖近红外定量模型中光谱范围选择方法的研究

Spectral range selection method in NIR quantitative model of tobacco water-soluble sugar

运用四种不同的光谱范围选择方法来建立烟草中水溶性糖的近红外定量模型，发现模型的交互验证系数、交互验证均方差和预测均方差有明显的差异。通过对烟草中水溶性糖的分子结构分析，结合傅里叶变换近红外漫反射光谱的特性，初步确定烟草水溶性糖近红外定量模型的建模光谱范围，以交互验证系数和交互验证均方差为评价指标进一步优化光谱范围，可以得到烟草水溶性糖在近红外定量模型中的最佳光谱范围为3850-5010 cm^-1、5720-7010 cm^-1和7760-7980 cm^-1，总糖和还原糖定量模型的交互验证系数、交互验证均方差和预测均方差分别为0.989、0.787、0.565和0.982、0.801、0.693。

Four different spectral range selection methods were applied in NIR quantitative model of tobacco water-soluble sugar. It was found that correlation coefficients of cross validation（R2）, root mean square errors of cross validation （RMSECV） and root mean square errors of prediction （RMSEP） were significantly different. Spectral ranges were preliminarily determined by analyzing molecular structures of tobacco water-soluble sugar and characteristic of FT-NIR diffuse reflectance spectroscopy. Two evaluating indicators, R2 and RMSECV, were used to further optimize spectral range. It was found that the optimal spectral ranges of tobacco water-soluble sugar was 3850~5010 cm^-1, 5720~7010 cm^-1and7760~7980 cm^-1. The final quantitative models of total sugar and reduced sugar, R2, RMSECV and RMSEP were 0.989, 0.787, 0.565 and 0.982, 0.801, 0.693, respectively.