RSRT-DETR: Hierarchical Polar Attention and Multiscale Hypergraph Networks for Dense Remote Sensing Small-Object Detection

Recent advances in optical remote sensing have expanded the use of uncrewed aerial vehicles (UAVs), yet accurately localizing dense, small objects in low-resolution UAV imagery remains challenging. We propose remote sensing real-time detection transformer (RSRT-DETR), a lightweight DETR derivative tailored for UAV-borne remote sensing that jointly enhances global context modeling and fine-grained detail preservation. RSRT-DETR couples a polar dynamic spatial finetuning (PDSF) attention module with a lightweight backbone to amplify fine-grained geometry while suppressing background noise, and embeds a multiscale hypergraph feature network (MS-HFNet) that reasons over high-order spatial relations to disambiguate cluttered targets. These features are hierarchically fused by the hypersemantic fusion network (HSFN), whose progressive edge-enhancement, resolution-adaptation, and multigranularity aggregation schemes jointly preserve cross-scale context without sacrificing speed. Evaluated on five challenging benchmarks-DOTAv1.0, VisDrone2019, NWPU VHR-10, DIOR, and HIT-UAV-RSRT-DETR achieves superior performance, with its AP50 surpassing state-of-the-art methods and its APS reaching the highest value. Meanwhile, the number of parameters and FLOPs is lower than those of similar DETR variants. The above results prove that RSRT-DETR achieves the optimal coordination of detection accuracy, model scale, and real-time performance in complex backgrounds and multiscale dense smalltarget scenes, providing a new paradigm for building efficient and reliable UAV remote sensing target detection systems. Our code is available on: https://github.com/zhy1109/RSRT-DETR