This paper described the development of a multivariate classification methodology to detect frauds in bovine meat based on mid-infrared spectroscopy and partial least squares discriminant analysis (PLS-DA). These frauds consisted of adding carrageenan, sodium chloride, and tripolyphosphate, ingredients that increase meat water holding capacity aiming to obtain economic gains. Meat pieces (fresh beef muscle) of the same bovine cut, M. semitendinosus, from different origins were injected with single to ternary mixtures of adulterants, and their purges were analyzed totaling 176 spectra. Multiclass PLS-DA models for specifically detecting each adulterant provided good results (correctly classification rates > 90%) only for tripolyphosphate. Nevertheless, a two-class PLS-DA model discriminating adulterated and non-adulterated meat provided high success rates (≥ 95%). Aiming to verify the model’s ability to detect other (non-trained) adulterant, this last model was combined with outlier detection in a soft version of a discriminant model that was able to correctly detect 100% of a new validation set consisting of 20 meat samples containing maltodextrin.

本文介绍了基于中红外光谱和偏最小二乘判别分析（PLS-DA）的多变量分类方法的开发，以检测牛肉中的欺诈行为。这些欺诈行为包括添加角叉菜胶，氯化钠和三聚磷酸盐，这些成分增加了肉类的持水量，旨在获得经济利益。相同的牛切肉，半腱肌肉的肉块（鲜牛肉肌）从不同来源注入了掺假品的单一或三元混合物，并分析了其净化物，共176个光谱。专门检测每种掺杂物的多类PLS-DA模型仅对三聚磷酸盐提供了良好的结果（正确分类率> 90％）。尽管如此，区分伪造和非伪造肉类的两类PLS-DA模型提供了很高的成功率（≥95％）。为了验证该模型检测其他（未经训练的）掺假物的能力，该最后一个模型与判别模型的软版本中的离群值检测相结合，该判别模型能够正确检测100％的新验证集（包括20个肉样品）含有麦芽糊精。