Determinação simultânea de adulteração em leite de cabra por adição de leite de vaca e teor de lipídios usando NIR portátio e algoritimos PLS

Abstract

The demand for goat's milk has been growing in recent years because it is a nutritionally rich food. In order to enjoy all the benefits offered and better use of nutrients, such as the lipids that play a regulatory role in goat milk in the consumer's body, the supplier must ensure that its original characteristics are preserved so that the product's quality level is maintained adequate. Thus, the present work aims to compare the performance of chemometric PLS regression models used to experiment with the percentage of cow's milk mixture in goat's milk and the lipid concentration using portable NIR spectrometry. The pre-processing used in the spectral data were SNV and MSC for scattering correction, baseline and Savitzky Golay derivation for baseline correction. For the construction of the calibration model, the partial least squares method (PLS) was used. Based on the results after applying the pre-processing, the best models obtained for quantifying the percentage of mixture were the models using Savitzky Golay derivation with 2nd derivative, 15-point window and 2nd order polynomial and Savitzky Golay with 2nd derivative, window 23 and 2nd order polynomial. The figures of merit evaluated were: RMSEC (10.17 and 12.19%), RMSECV (14.14 and 12.19%), RMSEP (10.66 and 10.04%), r2cal (0.89 and 0 .84 ), r2val (0.79 and 0.78) and r2pred (0.72 and 0.75) with the 15 and 8 latent variables, respectively. For lipid concentration, SNV showed the best results, obtaining the following values for RMSEC (0.28 g/100g), RMSECV (0.33 g/100g), RMSEP (0.27 g/100g), r2cal (0. 46) , r2val (0.33) and r2pred (0.27) using 7 latent variables. It is possible to observe values of figures of merit suitable for the proportion of bovine milk in goat milk. In terms of the model for lipid naivety, it was not possible to obtain adequate results to be used in real applications.