Atlantic salmon adulteration authentication by machine learning using bioimpedance non-destructive flexible sensing

Abstract

Temperature fluctuations at different stages of the supply chain increase the frozen-thawed cycle of perishable foods, potentially leading to quality and safety issues. For raw edible salmon in particular, it is not possible to ignore the issue of adulteration when frozen-thawed flesh is sold as fresh flesh. It is a challenge to achieve real-time detection of frozen-thawed salmon adulteration in fresh salmon. Existing impedance change ratio (Q-value) and PCA models cannot accurately authenticate frozen-thawed cycle adulterated salmon. In this paper, a flexible bioimpedance based non-destructive detection system was designed to authenticate adulterated salmon by online monitoring of changes in bioimpedance signals, ambient temperature, and relative humidity. The system provided a high level of monitoring accuracy and stability. Furthermore, an improved machine learning classification model based on principal component analysis - Bayesian optimization algorithm - support vector machine (PCA-BOA-SVM) was developed to effectively identify frozen-thawed adulterated salmon. The optimised model performance enhanced with prediction accuracy, precision, recall and F1 score of 0.9683, 0.9708, 0.9683 and 0.9679, respectively. This work could provide an effective solution to improve the authentication of food adulteration in the perishable food supply chain by improving traceability at all stages of the supply chain and sustainability of food industry development.