RSP-YOLOv11n:无人机航拍图像小目标检测模型OA

To address the challenges of small object detection in UAV aerial images,including insufficient feature extraction,in-formation loss in deep feature maps,high computational complexity,and inadequate real-time performance.we propose an im-proved model based on the YOLOv11n framework,named RSP-YOLOv11n.The model introduces the RFCBAMConv module,which combines channel attention and spatial attention mechanisms to enhance feature extraction for small objects.By dynami-cally adjusting the receptive field,the module overcomes the limitations caused by parameter sharing in standard convolution,thereby improving the flexibility and accuracy of feature extraction.In addition,the original SPPF layer is replaced with the PDC module,which employs convolution layers with different dilation rates for multi-scale feature extraction,effectively reducing de-tail loss while improving detection performance and computational efficiency for objects of different sizes.Furthermore,the pro-posed SEFP module uses P2-layer features,extracts small-object information through SPDConv,and fuses them with P3-layer features.Combined with CSP and OmniKernel,this module significantly improves the detection accuracy of small objects while avoiding the increase in computational cost and post-processing time caused by the conventional addition of a P2 layer.Experi-mental results on the VisDrone dataset show that,compared with the baseline model,RSP-YOLOv11n improves mAP@0.5 and mAP@0.5:0.95 by 6.7 percentage points and 4.5 percentage points,respectively.These results demonstrate its effectiveness and practicality in complex scenarios,as well as its ability to substantially enhance small object detection while maintaining rela-tively low computational cost.