AUTHENTICATION OF MEAT PRODUCTS FROM EXTENSIVE REARING SYSTEM BY NIR SPECTROSCOPY, HYPERSPECTRAL IMAGING, AND CHEMOMETRICS

Rearing system, breed, geographical origin are some of the quality features able to drive consumer’s choice in meat and meat products purchase. The same aspects are marketing leverage from the producer’s side. However, the compliance of a specific product with the declared - or supposed - quality features is mostly of a documentary nature, resulting in a possible distrust. Indeed, according to the Alert and Cooperation Network, meat and meat products are the third most reported category concerning non-compliance and suspicions fraud notifications, mostly related to faulty labelling or claims [1]. Bearing this framework in mind, both authorities and consumers could profit from the development of rapid, non-destructive solutions to certify meat and meat products with claims of quality. In this context, NIR spectroscopy could represents an efficient analytical solution to evaluate the authenticity of products. As a first step of this study, fat samples (n = 61) from extensively and intensively reared pigs were collected and analysed by NIR spectroscopy using a benchtop instrument. After an exploration of the dataset, an authenticity model for the “extensive” class was developed using DD-SIMCA [2]. The results showed that a model based on 4 PCs preprocessed using SNV allowed to reach perfect sensitivity and specificity by using only the fat portion in unprocessed meats [3]. However, when moving to processed products, minced and fermented such as ripened sausages, the things start to get complicated. Indeed, these products have usually highly non-uniform surfaces, with both fat and lean portions irregularly mixed and the analysis of the fat portion by single spot NIRS is often technically impossible. In the attempt to overcome this issue, as a second step of the work, the applicability of NIR HyperSpectral Imaging (HSI) to authenticate more complex products such as fermented cured meats was evaluated. Samples (n = 12) from extensive and intensive farming were collected and hyperspectral images of samples slices collected. Three distinct images of the samples were acquired. After background removal and data exploration, K-means clustering was used to separate the pixels of each sample into two clusters, corresponding to the fat and lean fraction. Subsequently, the NIR signal of the fat fraction of each salami was averaged, obtaining an average spectrum for each sample (12 x 140 matrix). The same approach was followed for the three images, obtaining a final matrix of 36 x 140. Preliminarily, an exploratory analysis was carried out by PCA (Principal Component Analysis). By means of PCA, a distribution of samples along PC1 according to the rearing system was highlighted in the score plot. Subsequently, SIMCA and PLS-DA models were built, showing promising errors in cross-validation, suggesting the potential of this technology. Nonetheless, the need to improve the numerosity of the dataset emerged. [1] European Union, 2023, available at https://food.ec.europa.eu/document/download/911d49f2-b3ef-4752-8ea3-5f20dbbe9945_ en?filename=acn_annual-report_2023.pdf. [2] Y.V. Zontov, O.Y. Rodionova, S.V. Kucheryavskiy, A.L. Pomerantsev, Chemometrics and Intelligent Laboratory Systems, 2017, 167, 23-28 [3] M.P. Totaro, G. Squeo, D. De Angelis, A. Pasqualone, F. Caponio, C. Summo, Journal of Food Composition and Analysis, 2023, 118, 105211