Novas estratégias analíticas baseadas em espectroscopia no infravermelho próximo e imagens digitais para identificação e quantificação de adulterações em leite caprino

Abstract

Adulteration of goat milk by adding cow milk has been a common fraudulent practice that can result in economic losses and possible damage to consumer health. Thus, this work aimed to develop green analytical methodologies based on Near-Infrared Spectroscopy (NIRS) and Digital Images for the identification and/or quantification of this type of adulteration. For this, a conventional benchtop NIR spectrophotometer was used, which was able to satisfactorily quantify both the adulteration and the fat content when coupled with the Successive

Projection Algorithm for selection of intervals in Partial Least Squares Regression (iSPA- PLS), while PLS demonstrated the best results for the protein content. The prediction results

were suitable for this purpose, since they achieved high correlation coefficients and low values of RMSEP (root mean square error of prediction) and REP (relative error of prediction), with RPD (ratio performance deviation) values higher than 3. However, considering that this type of adulteration does not depend on the amount of cow milk added to be characterized as a fraud, three different methodologies were developed for its rapid identification. In the first, using the same benchtop NIR equipment and PLS for Discriminant Analysis (PLS-DA), it was possible to identify additions of cow milk higher than 1% (m m-1 ), correctly classifying all pure and adulterated samples into their respective classes. Then, in order to make this methodology simple and portable, the suitability of using a miniaturized NIR and Digital Imaging as alternative low-cost analytical tools for the in situ detection of this type of adulteration was verified. In the case of the miniaturized NIR, the iSPA-PLS-DA attained the best predictive ability, correctly classifying a 100% of the pure goat milk samples and misclassifying only 1 adulterated sample as a pure one. Regarding the use of Digital Images, the best classification performance was obtained using the RGB color histogram and Data Driven-Soft Independent Modeling of Class Analogy (DD-SIMCA), reaching 90% of efficiency for discriminating the samples. Thus, considering both the cost, accessibility, and discriminant ability of the proposed methodologies, such techniques have proven to be adequate as quick and non-destructive tools for screening the authenticity of goat milk samples, with the potential to be applied in cooperatives and small industries.