LAUSR

Abstract A common fraudulent practice throughout the supply chain is the partial substitution of meat products with other meat species than the ones claimed. Thus, detection of meat adulteration using rapid, non-invasive methods during production, processing, distribution, consumption, and disposal of food products is of vital importance for the protection of consumers. In that framework the objective of this work was the detection of minced pork with chicken (and vice versa) adulteration in (a) raw fresh, (b) raw stored and (c) cooked minced meat using multispectral imaging (MSI) in tandem with support vector machines classification (SVM). Multispectral images were acquired at 0 h (freshly ground meat samples, n = 360), after 24 h and 48 h (stored meat samples, n = 180). Subsequently, at 48 h of storage, samples were cooked, and new images were acquired (n = 180). SVM classification models were developed and externally validated with sets from (a), (b) and (c). Their performance for the external validation in terms of accuracy were: for (a) freshly ground meat, 96.67% (n = 360), (b) stored minced meat, 0 h = 93.33%, 24 h = 97.78% and 48 h = 95.56%, (n = 180) and (c) cooked minced meat, 95.56% (n = 180). MSI data coupled with SVM algorithm exhibit promising potential for the rapid detection of adulterated meat samples in all three cases.