A Lightweight Model for UAV Detection Based on Improved YOLOv8

Abstract

UAV object detection under complex real-world conditions plays an important role in intelligent surveillance and airspace management. To overcome the limitations of existing YOLO-based approaches, particularly inadequate small-object detection performance and excessive model complexity in cluttered environments, a lightweight UAV detection framework named YOLOv8-CNM is proposed. The model improves the detection capability of small UAV targets while reducing computational burden. In the proposed framework, the original C2f module is replaced with the C2fCIB module to simplify the network structure and improve feature extraction efficiency. Furthermore, the NAM attention mechanism is integrated to enhance target-related feature representation and alleviate background interference, thereby strengthening the discrimination between UAV targets and complex backgrounds. Experimental evaluations demonstrate that YOLOv8-CNM achieves reliable detection accuracy while maintaining a lightweight architecture. Compared with the baseline model, the parameter count and FLOPs are reduced by 22.6% and 19.8%, respectively, whereas the mAP@0.5 increases by 0.3 percentage points. These findings indicate that the proposed method has strong potential for practical edge-device deployment in UAV detection tasks.