WR-YOLO: An Efficient and Lightweight Real-Time Detection Algorithm for Individual Dishes in Smart Canteens

The demand for intelligent management in canteens within the realm of smart retail is gradually increasing, placing higher requirements on the precise detection of dishes. This paper primarily addresses the task of single-category dish object detection, focusing on the construction of a single-category dish dataset and proposing a lightweight improved algorithm model. By integrating the lightweight backbone network PPLCNet, the attention mechanism MCA, and the bounding box loss MPDIoU, the model achieves enhanced detection accuracy while maintaining its lightweight nature. Experimental results demonstrate that the improved model outperforms the YOLOv8 baseline model on small-scale datasets, showing improvements in both mAP. 5 and mAP.95, and is more suitable for deployment on edge devices. The experiments indicate that this method holds significant potential for practical applications in real-world scenarios.