Deep learning in food quality and safety management: A review of architectures, applications, and future directions

As a cornerstone of the modern food industry, process control in food processing operations undergirds food quality assurance, safety compliance and economic performance of an enterprise. With artificial intelligence technology further penetrating into the food sector, deep learning has evolved from the early basic models such as Convolutional Neural Network (CNN), Recurrent Neural Network (RNN), and Artificial Neural Network (ANN) to a diversified repertoire of advanced architectures like YOLO, U-Net, Bidirectional Long Short-Term Memory neural network (BiLSTM) and Vision Transformer (ViT). These advanced architectures have been systematically investigated and deployed across multiple facets of process control in food processing stages. This review summarizes the multi-domain applications of deep learning in food process control, encompassing raw material inspection, dynamic monitoring, safety traceability, and quality evaluation. It further discusses diverse architectural strategies, including single-model and multi-model fusion approaches, and explores their integration with other intelligent analytical technologies to enhance processing efficiency and product consistency. Finally, current limitations regarding model interpretability and generalization are identified, and future research directions are proposed, focusing on the development of rapid, adaptive, and explainable deep learning systems to advance intelligent transformation across the entire food industry chain.