Real-Time Small UAV Detection in Complex Airspace Using YOLOv11 with Residual Attention and High-Resolution Feature...

Area of Science:

• Computer Vision
• Artificial Intelligence
• Aerospace Engineering

Background:

• Detecting small unmanned aerial vehicles (UAVs) in complex airspace is challenging due to their small size, resemblance to birds, and frequent occlusion.
• Existing detection methods struggle to accurately identify small UAVs in dynamic and cluttered environments.

Purpose of the Study:

• To develop a novel real-time detection framework for accurately identifying small UAVs in complex airspace.
• To enhance the YOLOv11 architecture with improved feature refinement and spatial detail preservation for small-object detection.

Main Methods:

• Integration of a Residual Convolutional Block Attention Module (ResCBAM) for enhanced feature refinement and residual connections.
• Incorporation of a high-resolution P2 detection head to maintain fine spatial details crucial for small-object localization.
• Evaluation on a custom dataset and cross-dataset validation on VisDrone2019-DET and UAVDT benchmarks.

Main Results:

• The YOLOv11-ResCBAM model achieved a mean average precision (mAP@0.5-0.95) of 0.845, a 7.9% improvement over the baseline YOLOv11n.
• Real-time inference speed of 50.51 frames per second (FPS) was maintained.
• Promising generalization capabilities were demonstrated through cross-dataset validation.

Conclusions:

• The proposed YOLOv11-ResCBAM framework effectively balances detection accuracy and computational efficiency for UAV surveillance.
• The integration of ResCBAM and the P2 detection head significantly improves the detection of small UAVs in challenging conditions.
• The model shows strong potential for deployment in security-critical environments requiring robust aerial surveillance.