基于D-Fine的棒棒糖微缺陷检测算法OA

To address the challenges in candy defect detection,where traditional models suffer from insufficient accuracy,large parameter counts,and slow inference speeds that fail to meet the real-time quality control requirements of high-speed production lines,an LRF-DFine detection algorithm integrating three core innovative modules is proposed,based on the D-Fine model.While preserving the original structure and foundation of the D-Fine model,the RepGhostCSPELAN lightweight module is incorporated to simplify the model and improve inference speed.The FMFFN mid-frequency feature enhancement module and the SimAM attention optimization module are utilized to suppress background interference and highlight critical regions.Experimental results show that,compared to the overall detection performance of the D-Fine model,the LRF-DFine model achieves comprehensive optimization in detection performance.Precision and recall are improved by 3.6 and 3.4 percentage points,respectively;mAP@0.5 is increased by 3.3 percentage points,and mAP@0.5∶0.95 is increased by 2.0 percentage points.The parameter count is reduced by 35.3%,computational complexity is decreased by 27.2%,and inference speed is increased by 20.0%.This research provides an effective technical solution for automated quality control in the food industry.