【摘要】 针对遥感图像中小目标检测面临的密集分布、尺寸微小、特征微弱易淹没及无人机视角下遮挡严重等问题，提出改进算法RSA-YOLOv11。首先，在骨干网络引入RFAConv模块，扩大感受野，强化小目标特征提取能力；其次，引入语义与细节注入SDI模块，自适应增强关键特征并抑制冗余信息；最后，新增P2检测层，结合自适应空间特征融合ASFF模块重构头部网络。结果表明，在VisDrone数据集上和YOLOv11n相比，RSA-YOLOv11模型在mAP@0.5和mAP@0.5∶0.95指标上分别提升7.6%和5.3%,表现优越。更多还原

【Abstract】 To address challenges like dense distribution, tiny size, weak features, and severe occlusion in drone-based remote sensing imagery, we propose RSA-YOLOv11. Key improvements include: 1) Integrating RFAConv in the backbone to expand receptive fields and enhance feature extraction. 2) Adding an SDI module to adaptively highlight critical features while suppressing noise. 3) Introducing a P2 detection layer with ASFF to reconstruct the head network for optimized feature fusion. On the VisDrone dataset, RSA-YOLOv11 outperforms YOLOv11n with 7.6% and 5.3% gains in mAP@0.5 and mAP@0.5∶0.95 respectively.更多还原