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

The rapid development of unmanned aerial vehicles (UAVs) has made object detection in aerial imagery a key focus in computer vision. However, challenges like small object sizes, complex backgrounds, and lighting variations hinder detection accuracy. This paper presents UAV-Det, an efficient multi-scale detection framework for UAV imagery. UAV-Det consists of three main components: a Multi-Scale Fusion Perception (MSFP) module for enhanced perception and localization, an Adaptive Attention Feature Optimizer (AAFO) for focusing on target features while minimizing background noise, and a Small-Object Guided Edge Enhancement (SGEE) module for improving small-object detection. Experiments on standard aerial datasets show that UAV-Det achieves an mAP of 92.7%, surpassing mainstream detection algorithms and demonstrating strong robustness in complex environments.