改进RT-DETR的输电线路异物检测算法研究OA

To address challenges including limited detection accuracy,high computational complexity,and feature extrac-tion difficulties in UAV aerial imagery,an improved RT-DETR algorithm is proposed.A lightweight DynRepFusion block(DRF block)in the backbone network enhances detection accuracy while significantly reducing model complexity and computational costs.The dynamic feature region collaborative attention(DFRCA)module uses dual-path histogram reor-ganization to mitigate false detection rates for dense targets.The multi-scale feature fusion network(MSFFN)achieves synchronous optimization across multiple scales.The EIoU loss function improves detection robustness against image size variations.Experimental results show that the improved model reduces parameters by 26.1％and GFLOPs by 22.2％,while increasing mAP50 and mAP50:95 to 94.5％and 76.2％,with 4.2 and 2.7 percentage points of improvement over the original model,respectively.Compared with YOLOv8,which has the best comprehensive performance among the mainstream algorithms,the proposed model improves mAP50 and F1-score by 2.1 and 3.9 percentage points,respectively.The improved RT-DETR algorithm enhances detection accuracy,reduces the false detection rate and saves computational resources during UAV inspection tasks,providing an effective solution for UAV-based object detection.