YOLO-DD: a lightweight framework for UAV detection in complex environments via boundary-aware fusion

Abstract

The rapid proliferation of unmanned aerial vehicles (UAVs) poses significant challenges for collision avoidance and privacy protection, particularly due to the high missing detection rate of small UAV targets in complex backgrounds and multi-scale scenarios. To address these challenges, this study proposes YOLO-DD, an enhanced lightweight detection model based on YOLOv11n. First, a Ghost dynamic C3K2 (GD-C3K2) module is designed by integrating Ghost convolution and dynamic convolution, effectively reducing computational complexity while enhancing multi-scale feature extraction. Second, a Selective Boundary Aggregation module is introduced to optimize cross-scale feature fusion and refine localization accuracy for small targets. Finally, a Lightweight Shared Convolutional Detection head is implemented to balance accuracy and efficiency through parameter-sharing mechanisms. Comprehensive experiments on the Det-Fly dataset demonstrate YOLO-DD’s superiority over state-of-the-art methods: achieving 97.4% mAP50 (14.0% higher than baseline YOLOv11n, 3.8% higher than YOLOv10), while maintaining the smallest model size with only 1.8 M parameters (5.3% fewer than YOLOv10, 30.8% fewer than YOLOv11) and lowest computational cost of 5.6 GFLOPs. Despite its compact design, YOLO-DD delivers real-time performance at 87.72 FPS, outperforming most YOLO variants in both accuracy and efficiency. The proposed model significantly improves detection robustness in cluttered environments, offering a high-precision, lightweight solution for UAV surveillance and emergency response applications.