Related Experiment Videos
AQF-Net: Adaptive query modeling and efficient feature fusion for UAV tiny-object detection
Yong He1, Yifan Tang1, Renfeng Xiao1
1College of Artificial Intelligence, Changsha University of Science and Technology, Changsha, China.
Plos One
|August 13, 2026
Summary
This study introduces AQF-Net, a new framework for detecting tiny objects in drone imagery. AQF-Net improves accuracy by efficiently modeling context and fusing multi-scale features for complex aerial scenes.
Area of Science:
- Computer Vision
- Artificial Intelligence
- Remote Sensing
Background:
- Tiny-object detection in Unmanned Aerial Vehicle (UAV) aerial imagery is difficult due to small object scales, dense arrangements, and complex backgrounds.
- Existing methods struggle with inefficient query modeling and insufficient multi-scale feature representation, especially in high-resolution images with varying target densities.
Purpose of the Study:
- To propose AQF-Net, a unified detection framework designed to overcome the limitations of current methods for tiny-object detection in UAV imagery.
- To enhance the model's ability to adapt to complex UAV scenarios by optimizing feature representation and query generation.
Main Methods:
- AQF-Net is built upon the D-FINE architecture and integrates three key components: Fixed-Query Self-Attention (FQSA) for global context modeling, Large-Receptive-Field Enhancement (LREA) for multi-scale feature fusion, and adaptive query modeling for density-aware allocation.
- These components work together to jointly optimize feature representation and query generation.
Main Results:
- AQF-Net demonstrated superior performance compared to the D-FINE baseline and other state-of-the-art methods on the CODrone, VisDrone2019, and PV-DV datasets.
- The model achieved 33.4% AP and 55.0% AP50 on the VisDrone2019 validation set, showcasing improved tiny-object detection capabilities.
- AQF-Net maintained a strong balance between detection accuracy and computational efficiency.
Conclusions:
- AQF-Net effectively addresses the challenges of tiny-object detection in complex UAV aerial imagery.
- The proposed framework offers a promising solution for improving the accuracy and efficiency of object detection in high-resolution aerial scenes.