本研究旨在探讨利用近红外光谱技术评估高粱中粗蛋白质、水分含量的可行性。以收集的110份高粱样品作为研究对象，采用GB/T 6432-1994、GB/T 6435-2014中方法分别对粗蛋白质、水分含量进行测定，利用傅里叶变换近红外光谱仪采集样品的近红外漫反射光谱，光谱扫描范围4 000~12 800 cm^-1，分辨率16 cm^-1，样品重复装样扫描4次，每次扫描64次获得平均光谱，取4次扫描光谱作为样本的原始光谱。分别选取矢量归一化、最小-最大归一化、一阶导数、二阶导数、多元散射校正、一阶导数+减去一条直线、一阶导数+矢量归一化、一阶导数+多元散射校正探索适用于高粱中粗蛋白质、水分含量的光谱预处理方法。利用定标集样品光谱数据，采用偏最小二乘方法结合全交互验证手段来防止过拟合现象，建立定标模型。在此基础上，利用定标决定系数、定标均分根误差、定标相对分析误差、交互验证决定系数、交互验证均方根误差、交互验证相对分析误差确定最优模型。结果显示：粗蛋白质含量扫描光谱采用一阶导数+多元散射校正光谱预处理，光谱范围为9 401.9~5 443.6 cm^-1与4 603.0~4 243.9 cm^-1。水分含量扫描光谱采用一阶导数+减去一条直线，光谱范围为7 500.3~6 096.5 cm^-1与5 451.8~4 243.9 cm^-1。高粱中粗蛋白质、水分含量的近红外光谱预测模型定标相对分析误差分别为8.41、12.20；交互验证相对分析误差分别为4.97、7.97；外部验证相对分析误差分别为3.32、5.36。由结果可知，本研究建立的高粱中粗蛋白质和水分含量的近红外光谱预测模型的相对分析误差均大于评估值，具有精确地评估高粱中粗蛋白质和水分含量的应用效果。

The objective of the present study was to explore the possibility of evaluation of crude protein and moisture contents in sorghum grain using by near-infrared spectroscopy technique. One hundred and ten sorghum grain samples were collected in the experiment. The contents of crude protein and moisture in sorghum grain samples were determined by GB/T 6432-2018 and GB/T 6435-2014, respectively. Near infrared diffuse reflectance spectra of samples were collected by four times transform near infrared spectrometer. Spectral scanning ranges was from 4 000 to 12 800 cm^-1 with resolution of 16 cm^-1. Each sample was scanned for 4 times repeatedly. The average spectra were obtained by scanned 64 times. Vector normalization, minimum-maximum normalization, first derivative, second derivative, multivariate scattering correction, first derivative+minus a straight line, first derivative+vector normalization and first derivative+multivariate scattering correction were selected to explore the spectral pretreatment method suitable for the contents of crude protein and moisture in sorghum grain. The calibration models were established using spectral data of samples in calibration set, and the combination of partial least squares combined with whole cross-validation method was use to prevent overfitting. On that basis, the optimal model was determined according to coefficient of determination for calibration, root mean square error of calibration, residual predictive deviation of calibration, coefficient of determination for cross-validation, root mean square error of cross-validation and residual predictive deviation of cross-validation. The results showed that first derivative+multivariate scattering correction as optimal spectral pretreatment was used for crude protein content in sorghum grain, and the spectral ranges were 9 401.9 to 5 443.6 cm^-1 and 4 603.0 to 4 243.9 cm^-1. First derivative+minus a straight line as optimal spectral pretreatment was fit for moisture content in sorghum grain, and the spectral ranges were 7 500.3 to 6 096.5 cm^-1 and 5 451.8 to 4 243.9 cm^-1. The residual predictive deviations of calibration of near-infrared spectroscopy prediction model for the contents of crude protein and moisture in sorghum grain were 8.41 and 12.20, respectively. The residual predictive deviations of cross-validation of near-infrared spectroscopy prediction model for the contents of crude protein and moisture in sorghum grain were 4.97 and 7.97, respectively. The residual predictive deviations of external validation of near-infrared spectroscopy prediction model for the contents of crude protein and moisture in sorghum grain were 3.32 and 5.36, respectively. The results indicate that the residual predictive deviations of the near-infrared spectroscopy prediction models for crude protein and moisture contents in sorghum grain which established in this study are bigger than assessed values, therefore they have application effect to accurately evaluate the contents of crude protein and moisture in sorghum grain.