Deep learning hyperspectral imaging: a rapid and reliable alternative to conventional techniques in the testing of food quality and safety

Abstract

Food quality and safety are a great public concern; outbreaks of food-borne illnesses can lead to different health problems. Consequently, rapid and non-destructive artificial intelligence approaches are required for sensing the safety situation of foods. As a promising technology, deep learning for hyperspectral imaging (HSI) has the potential for rapid food safety and quality evaluation and control. Spectral signatures of food substances are sensitive to water content variation, the extent of hydrogen bonding, geographical origin, harvesting time and the variety of food under study. Deep learning models have shown great potential in addressing the challenge of sensitivity of spectral signatures of food substances. After discussing the basics of HSI, this review provides a detailed study of various deep-learning algorithms that have been put to use via HSI in the determination of sensory and physicochemical properties, adulteration and microbiological contamination of food products. The existing literature includes HSI for evaluating quality attributes and safety of different food categories like fruits, vegetables, cereals, milk and meat. This paper presents a practical framework for deep learning-based food quality assessment using hyperspectral imagery. We demonstrate its versatility across diverse food quality domains and provide a concise step-by-step guide for researchers. It has been predicted that deep learning for HSI can be considered a reliable alternative technique to conventional methods in realising rapid and accurate inspection, for testing food quality and safety.