花生中蛋白质含量与分布能够显著影响花生制品品质。 利用高光谱图像结合化学计量学研究可视化花生中蛋白质含量分布的可行性。 从校正后的花生图像的感兴趣区域（region of interest, ROI）中提取光谱信息, 通过传统化学方法测定蛋白质含量。 比对了不同光谱预处理和回归算法, 以二阶导数（the second derivative, 2nd-der）为最佳的光谱预处理方法, 偏最小二乘法（partial least squares, PLS）为最佳的回归算法。 基于预处理后的光谱和花生蛋白质的化学值, 建立全波长PLS模型, 全波长模型具有良好的性能（校正集相关系数为0.91, 校正集标准偏差0.86; 预测集相关系数为0.86, 预测集标准偏差为0.69）。 利用回归系数法（regression coefficient, RC）从全波长模型中选择14个特征波长, 建立2nd-der-RC-PLS特征波长模型, 模型性能（校正集相关系数为0.86, 校正集标准偏差1.03; 预测集相关系数为0.80, 预测集标准偏差为0.77）与全波长模型相当。 采用2nd-der-RC-PLS算法将花生高光谱图像转变成蛋白质含量分布图。 成对t检验判断凯氏定氮法与高光谱法无显著性差异。 结果表明结合化学计量学的高光谱成像技术为测定花生中蛋白质含量分布提供了一种高效非破坏性方法。

The study aims to explore the potential of hyperspectral imaging (HSI) with chemometrics for rapidly and non-invasively visualizing the spatial distribution of protein content which can affect the quality of peanut products as a critical component of peanut. Spectral data contained in the region of interest (ROI) of the corrected hyperspectral images of peanut were extracted and protein contents were measured with conventional chemical method. By comparing different pretreatments and modeling algorithms, the second-order derivatives (2nd-der) on spectra is optimal pretreatment, and partial ceast square (PLS) is the best regression method. Based on the pretreatment spectra and the measured protein content model, a good performance model (RC=0.91, SEC=0.86; RP=0.86, SEP=0.69) was built with full wavelengths. The fourteen optimal wavelengths were carried out based on the regression coefficients (RC) of the established PLS model. Then, using optimal wavelengths built RC-PLS model which show resembling performance (RC=0.86, SEC=1.03; RP=0.80, SEP=0.77). At last, an imaging processing algorithm was developed to transfer each pixel in peanut to protein content with the 2nd-der-RC-PLS model. There was no significant difference between Kjeldahl and HSI method by the paired test. The result demonstrated the capacity of HSI in combination with chemometrics for fast and non- destructively determining protein content in peanut.