RFAG-YOLO: A Receptive Field Attention-Guided YOLO Network for Small-Object Detection in UAV Images

Area of Science:

• Computer Vision
• Object Detection
• UAV Imagery Analysis

Background:

• YOLO object detection methods are efficient but struggle with small objects in UAV images due to low resolution and scale variations.
• Challenges include degraded feature extraction and limited detection performance in complex environments.

Purpose of the Study:

• To develop an advanced YOLO adaptation for robust small-object detection in UAV imagery.
• To improve feature representation and detection accuracy under challenging conditions.

Main Methods:

• Proposed Receptive Field Attention-Guided YOLO (RFAG-YOLO), an adaptation of YOLOv8.
• Introduced a novel Receptive Field Network (RFN) block for fine-grained detail capture.
• Designed an enhanced FasterNet module and a Scale-Aware Feature Amalgamation (SAF) component.

Main Results:

• RFAG-YOLO outperformed YOLOv7, YOLOv8, YOLOv10, and YOLOv11 on the VisDrone2019 dataset.
• Achieved an mAP50 of 38.9%, with significant improvements over baseline models.
• Demonstrated high efficiency, achieving 97.98% of YOLOv8s performance with only 53.51% of its parameters.

Conclusions:

• RFAG-YOLO offers superior accuracy and efficiency for small-object detection in UAVs.
• The method is highly suitable for resource-constrained UAV applications.
• Shows significant potential for real-world applications requiring precise detection under challenging conditions.