SFA-DETR: An Efficient UAV Detection Algorithm with Joint Spatial-Frequency-Domain Awareness.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Signal Processing

Background:

• Unmanned Aerial Vehicle (UAV) detection is hindered by small target sizes, loss of detail, and complex backgrounds.
• Existing methods struggle to effectively integrate spatial and frequency domain information for robust UAV detection.

Purpose of the Study:

• To propose a novel object detection algorithm, Spatial-Frequency Aware DETR (SFA-DETR), for improved UAV detection.
• To enhance both spatial and frequency domain perception for more accurate and efficient UAV identification.

Main Methods:

• Developed IncepMix backbone for dynamic fusion of multi-scale spatial information, reducing computational cost.
• Introduced Frequency-Guided Attention Block (FGA Block) for enhanced boundary perception using frequency-aware guidance.
• Incorporated adaptive sparse attention mechanism to prioritize critical semantic features.

Main Results:

• SFA-DETR achieved a 1.2% improvement in mAP50 and a 1.7% improvement in mAP50:95 on the DUT Anti-UAV dataset.
• Reduced parameter count by 14.44% and computational cost by 3.34% compared to existing methods.
• Demonstrated a balance between detection accuracy and computational efficiency.

Conclusions:

• The proposed SFA-DETR algorithm effectively addresses challenges in UAV detection.
• The integration of spatial and frequency domain analysis leads to superior performance and efficiency.
• SFA-DETR offers a promising solution for real-world UAV detection applications.