Lightweight Fruit and Vegetable Detection Research Based on YOLO-PDMS

To reduce the memory footprint and computational requirements of the model, a lightweight model YoLo-PDMS was proposed. It introduced modules named P-ELAN and D-ELAN and replaced the original CIoU with MPDIoU. Network Slimming was utilized to reduce the model size and parameter count. In the final experiment, compared to the original model, parameters decreased by 90.1%, FLOPs reduced by 87.79%, latency decreased by 56.45%, FPS increased by 2.29 times, model size decreased by 77.78%, while MAP50 only decreased by 0.1%.