H-Yolov11n Model for Tiny Object Recognition in Aerial Images

To address the challenge of low accuracy in tiny target recognition within aerial imagery scenarios, this study proposes an H-YOLOv11n model with restructured network architecture. Building upon the YOLOv11 framework, the improved architecture incorporates a shallow-level P2 layer while eliminating the deep-level P5 layer, thereby transforming the original P3-P4-P5 detection layers of YOLOv11n into a novel P2-P3-P4 configuration. Experimental evaluations on the AI-TOD dataset demonstrate that the enhanced model achieves a mAP50 of 20.3% at <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$640 \times 640$</tex> resolution, representing a 1.6% improvement over the baseline YOLOv11n model while reducing parameters by 24.7%. Visual detection results substantiate the model's effectiveness, providing a viable lightweight solution for small target recognition in aerial imaging applications.