Research has shown that near-infrared hyperspectral imaging (NIR HSI) is an effective rapid-detection tool for milk powder adulteration, but model generality under different adulteration conditions requires further study. Therefore, this study focused on developing generic models for the detection of vanillin and melamine in infant formulas from diverse brands. Three pretreatment algorithms were applied successively to spectrum of each pixel in hyperspectral images. Minimum noise fraction was applied to eliminate interference from brand diversity and extract adulterant information. Partial least squares discriminant analysis (PLSDA) was used to develop a classification model to identify vanillin-rich pixels. The PLSDA model, developed with three optimal wavelengths selected by the successive projections algorithm (SPA), detected vanillin at concentrations as low as 0.01%. Partial least squares regression (PLSR) was applied to establish a quantitative model for melamine. The PLSR model, established with six optimal wavelengths selected by the competitive adaptive reweighting algorithm (CARS), showed excellent predictive capabilities, with a limit of detection of 0.5%. A visual prediction map clearly showed the location of vanillin-rich pixels and melamine content variations spatially. The proposed generic practical method would greatly facilitate the application and promotion of NIR HSI technology in quality inspection for the milk powder market and manufacturers.

研究表明，近红外高光谱成像（NIR HSI）是奶粉掺假的有效快速检测工具，但是在不同掺假条件下的模型通用性需要进一步研究。因此，本研究着重于开发通用模型以检测来自不同品牌的婴儿配方食品中的香兰素和三聚氰胺。三种预处理算法被依次应用于高光谱图像中每个像素的光谱。应用了最小噪声分数，以消除来自品牌多样性的干扰并提取掺假信息。偏最小二乘判别分析（PLSDA）用于开发分类模型，以识别富含香兰素的像素。PLSDA模型是根据连续投影算法（SPA）选择的三个最佳波长开发的，检出的香兰素浓度低至0.01％。应用偏最小二乘回归（PLSR）建立三聚氰胺定量模型。通过竞争性自适应重加权算法（CARS）选择了六个最佳波长建立的PLSR模型具有出色的预测能力，检出限为0.5％。视觉预测图清楚地显示了富含香兰素的像素的位置和三聚氰胺含量在空间上的变化。所提出的通用实用方法将极大地促进NIR HSI技术在奶粉市场和制造商的质量检验中的应用和推广。通过竞争性自适应重加权算法（CARS）选择的六个最佳波长建立的色谱柱具有出色的预测能力，检出限为0.5％。视觉预测图清楚地显示了富含香兰素的像素的位置和三聚氰胺含量在空间上的变化。所提出的通用实用方法将极大地促进NIR HSI技术在奶粉市场和制造商的质量检验中的应用和推广。通过竞争性自适应重加权算法（CARS）选择的六个最佳波长建立的色谱柱具有出色的预测能力，检出限为0.5％。视觉预测图清楚地显示了富含香兰素的像素的位置和三聚氰胺含量在空间上的变化。所提出的通用实用方法将极大地促进NIR HSI技术在奶粉市场和制造商的质量检验中的应用和推广。