Abstract: With its advantages of being non-contact, rapid, and automated, computer vision technology has provided efficient solutions for many fields traditionally reliant on manual judgment. In the domain of food recognition, conventional detection methods have long suffered from limitations such as low efficiency, strong subjectivity, and difficulties in quantification, making it challenging to address the complexity and diversity of food appearance characteristics. To address these issues, this paper systematically reviews the research progress, key applications, and future directions of computer vision in food recognition. It focuses on analyzing the key technical approaches and application effectiveness of deep learning models, represented by Convolutional Neural Networks and the YOLO series algorithms, in ingredient recognition, dish classification, and food quality inspection. The paper also outlines the evolution of mainstream datasets such as Food-101 and Nutrition5k from coarse-grained classification towards fine-grained and multi-modal correlation, highlighting the critical impact of dataset quality on model performance. On this basis, the paper further explores the practical value of this technology in typical scenarios such as nutrition and health management, intelligent catering settlement, and safety inspection. Finally, it provides an outlook on future research directions, pointing out that efforts should focus on cutting-edge pathways such as multi-modal information fusion, model lightweighting and edge deployment, and few-shot learning. These advancements are expected to promote the deeper role of food recognition technology in personalized nutrition, whole-chain safety supervision, and the construction of sustainable dietary systems, ultimately achieving the deep integration and systematic innovation of computer vision and food science.