Machine vision combined with deep learning–based approaches for food authentication: An integrative review and new insights

Abstract

Food fraud undermines consumer trust, creates economic risk, and jeopardizes human health. Therefore, it is essential to develop efficient technologies for rapid and reliable analysis of food quality and safety for food authentication. Machine vision–based methods have emerged as promising solutions for the rapid and nondestructive analysis of food authenticity and quality. The Industry 4.0 revolution has introduced new trends in this field, including the use of deep learning (DL), a subset of artificial intelligence, which demonstrates robust performance and generalization capabilities, effectively extracting features, and processing extensive data. This paper reviews recent advances in machine vision and various DL-based algorithms for food authentication, including DL and lightweight DL, used for food authenticity analysis such as adulteration identification, variety identification, freshness detection, and food quality identification by combining them with a machine vision system or with smartphones and portable devices. This review explores the limitations of machine vision and the challenges of DL, which include overfitting, interpretability, accessibility, data privacy, algorithmic bias, and design and deployment of lightweight DLs, and miniaturization of sensing devices. Finally, future developments and trends in this field are discussed, including the development of real-time detection systems that incorporate a combination of machine vision and DL methods and the expansion of databases. Overall, the combination of vision-based techniques and DL is expected to enable faster, more affordable, and more accurate food authentication methods.