YOLO-MAG: an edge-preserving and bilateral dual-gated YOLO11 network for small object detection in UAV aerial imagery

Abstract Small object detection in unmanned aerial vehicle (UAV) imagery is severely hindered by low spatial resolution, faint boundaries, drastic scale variations, and complex background interference. To address these vulnerabilities, we propose YOLO-MAG, an edge-preserving architecture built upon YOLO11 for aerial surveillance. First, the existing MEIS operator is integrated into the C3k2 framework and deployed across both the backbone and neck, forming a continuous edge-preserving feature path. Second, ACmix is embedded into the C2PSA block to combine global contextual modeling with local convolutional inductive bias. Third, the original single-gated GAC is reconstructed as a bilateral dual-gated module and incorporated into a modified feature pyramid network, in which the two feature branches jointly participate in estimating the modulation weights. These task-oriented architectural adaptations constitute the YOLO-MAG framework. Evaluated on the VisDrone2019 benchmark, YOLO-MAG-n achieves 40.4% [email protected] and 24.5% mAP@[0.5:0.95], outperforming the YOLO11n baseline by 6.9 and 5.0 percentage points, respectively, while sustaining real-time inference at 136 FPS on an NVIDIA RTX 4080 GPU. Additional evaluation on UAVDT demonstrates consistent performance gains on a second aerial traffic dataset.