基于YOLOv5的食物识别系统设计与实现

Design and Implementation of Food Recognition System Based on YOLOv5

摘要: 本研究提出了一种基于YOLOv5的智能食物识别系统。通过构建UNIMIB2016_YOLOv5数据集(含73类3616个标注实例)，提升了复杂场景下的识别鲁棒性。实验表明，YOLOv5s模型在保持轻量化的同时，实现了85.2%的mAP@0.5检测精度和63 FPS的实时推理速度。系统集成PyQT5可视化界面与MySQL数据库，支持端到端的图像/视频处理流水线，已成功应用于食物识别系统。

Abstract: This study presents an intelligent food recognition system based on YOLOv5. By constructing the UNIMIB2016_YOLOv5 dataset (containing 73 categories with 3616 annotated instances), the robustness of recognition in complex scenarios has been enhanced. Experimental results demonstrate that the YOLOv5s model achieves a detection accuracy of 85.2% mAP@0.5 while maintaining a lightweight structure, along with a real-time inference speed of 63 FPS. The system integrates a PyQT5-based graphical user interface and a MySQL database, supporting an end-to-end image/video processing pipeline. It has been successfully applied to food recognition systems.