Abstract:In unmanned aerial vehicle (UAV) countermeasure systems, UAV targets exhibit prominent multi-scale characteristics, particularly in small target detection, where detection accuracy is often lower. To address this issue, a YOLOv8n-based optimization algorithm, YOLOv8-C2f-RFCBAMConv, is proposed, which incorporates a dynamic sample attention scale sequence. The challenges of multi-scale and small target recognition in UAVs are analyzed, and it is suggested that the feature extraction capability can be optimized by improving the backbone network and integrating the C2f-RFCBAMConv module. The RFCBAM mechanism, combined with residual fusion and context attention mechanisms, enhances feature representation while reducing computational complexity. Furthermore, the WIoU loss function is employed to mitigate the impact of low-quality data from small targets on gradient propagation, accelerating network convergence. Experimental results demonstrate that the proposed model achieves a 3.1% and 1.7% improvement in mAP@0.5 and mAP@0.5:0.95, respectively, on a self-collected UAV dataset, with a 0.7 GFLOP reduction, indicating higher detection accuracy and lower computational complexity.
