基于跨尺度协同感知的无人机目标检测算法OA

To address the challenges of dense distribution of small targets and significant multi-scale variations in unmanned aerial vehicle(UAV)aerial images,this paper proposes a Cross-Scale Collaborative Perception YOLO(CSCP-YOLO)object de-tection algorithm.Firstly,a cross-scale adaptive perception module is designed,which achieves adaptive perception of targets at different scales through the collaborative fusion of receptive field attention mechanism and coordinate attention mechanism.Second-ly,a cross-scale collaborative feature fusion network is constructed,which establishes a four-level cross-scale perception pyramid from P2 to P5 through sequential collaborative fusion and feature collaborative enhancement.Finally,a cross-scale collaborative perception detection head is proposed,which adopts a cross-scale parameter collaborative sharing strategy to achieve unified detec-tion of multi-scale targets.Experimental results on the VisDrone2019 dataset show that CSCP-YOLO achieves 6.9%improvement in mAP@0.5 and 4.3%improvement in mAP@0.5:0.95 compared to the baseline YOLOv11n algorithm,while reducing the parame-ter count by 5.5%.Generalization experiments on the DOTA v1.0 dataset further validate the cross-scale perception capability of the algorithm.