AeroDet-Net: A Deep Learning-Based Object Detection Algorithm for UAV Aerial Imagery

Object detection in UAV aerial imagery is challenged by small object scales, dense distributions, and complex backgrounds, which hinder accurate real-time detection. Although YOLOv12 performs well in general tasks, its robustness and small-object detection capability in aerial scenarios remain limited. To address these challenges, this paper proposes AeroDet-Net, an enhanced framework based on YOLOv12. The model incorporates a Multi-Scale Aerial Feature Fusion (MSAFF) module to improve multi-scale feature interaction, a Spatial-Context Guided Attention (SCGA) mechanism to strengthen target-focused responses while suppressing background interference, and a Small-Object Micro Enhancement (SOME) unit to enhance small-object representation. Experiments on a public UAV dataset demonstrate that AeroDet-Net achieves a mean Average Precision (mAP) of 96.2% while maintaining real-time performance, outperforming the baseline YOLOv12 and showing strong effectiveness in complex aerial environments.