AEFFNet: Attention Enhanced Feature Fusion Network for Small Object Detection in UAV Imagery
Zhaoyu Nian, Yang Wenzhu, Hao Chen
Abstract
The rapid advancement of unmanned aerial vehicle (UAV) technology has markedly increased the use of drone-captured imagery across various applications, necessitating enhanced accuracy and real-time performance in UAV image detection. Addressing the specific challenges posed by small and densely distributed objects in such images, we introduce an attention enhanced feature fusion network (AEFFNet) designed specifically for small object detection in UAV imagery. Firstly, a hybrid attention module with associated multi-axis frequency and spatial attention was designed to enhance the feature extraction of small objects. Secondly, an adjacent layer feature fusion module is innovatively proposed in order to boost the detection capabilities for small and occluded objects. Finally, a series experiments are conducted on the VisDrone2023 dataset, which involves a large number of small objects photographed by drones. Our evaluations, conducted on the VisDrone2023 dataset, demonstrate substantial improvements over the YOLOv8m baseline model, with a 3.0% increase in mean Average Precision (mAP) and a 4.4% rise in AP50.