YOLO11n-SMSH: An Improved UAV Target Detection Model For YOLO11n

In response to the challenges of posture diversity, motion blur and small target detection in unmanned aerial vehicle (UAV) target detection in long-distance and complex scenarios, this paper proposes an improved UAV target detection model:YOLO11n-SMSH. This model significantly improves the detection performance through four core mechanisms: Firstly, a CRIE module with edge perception integration is introduced in the backbone network to enhance the ability of extracting target edge features; Secondly, a SRA-DFF network with semantic association enhancement capability is used as the neck network to achieve high-quality feature interaction and fusion; Furthermore, the NTTAA detection head is finely designed, and through the weight sharing mechanism and bidirectional parallel task alignment path, the collaboration between classification and localization tasks is effectively strengthened; Finally, the GIoU loss function is introduced, and the boundary box regression is optimized using spatial coverage, improving the model's adaptability to UAV targets. Experimental results on the DUT Anti-UAV dataset show that YOLO11n-SMSH performs excellently. Compared with the baseline model, the accuracy (P), recall rate (R), mAP50 and mAP50-95 have significantly increased by 1.5%, 3.9%, 2.6% and 2.5%, respectively. The experimental results verify the effectiveness of the multi-module collaborative optimization strategy and provide a high-performance solution for UAV target detection tasks in practical applications.