Small Targets Under Drone Perspective Based on Improved YOLOv8s Detection Algorithm

To address the challenges of micro-scale targets, severe occlusion, and uneven spatial distribution in UAV aerial imagery, this paper proposes an improved lightweight YOLOv8 model for Drone Perspective detection. Comparative experiments across modules demonstrate: recall rate increased by 2.4%, mAP surged to 42.5%—representing a 2.6% improvement over the baseline model—with significant gains in both recall and mAP compared to other detection models, establishing overall superiority. This research pioneers a novel pathway for micro-target detection, offering valuable practical guidance for algorithmic selection and refinement.