A frequency-saliency guided multi-scale feature fusion network for robust UAV object detection

Area of Science:

• Computer Vision
• Artificial Intelligence
• Remote Sensing

Background:

• Unmanned Aerial Vehicle (UAV)-based object detection is crucial for surveillance and traffic monitoring.
• Challenges include small object scales, complex backgrounds, dense distributions, and occlusions, hindering accuracy and real-time performance.

Purpose of the Study:

• To propose a frequency-saliency guided multi-scale feature fusion network based on YOLO11 for improved UAV object detection.
• To enhance the balance between detection accuracy and real-time inference speed.

Main Methods:

• Designed a Global Frequency Feature Extraction module using saliency enhancement and Dual-Tree Complex Wavelet Transform.
• Introduced a lightweight multi-scale feature fusion module with soft gating and spatial attention.
• Employed a Scale-Adaptive Dropout Feature Pyramid Network for robust multi-scale feature learning.

Main Results:

• The proposed FSMF-YOLO achieved 91.6% precision, 83.5% recall, and 89.3% mAP@50 at 101 FPS on the HIT-UAV dataset.
• Demonstrated a 2.6% improvement in mAP@50 over YOLO11n while maintaining real-time inference.
• Showcased improved small-object representation in complex UAV scenes.

Conclusions:

• Frequency-saliency guidance effectively enhances small-object detection in challenging UAV environments.
• The proposed network offers a favorable accuracy-efficiency trade-off compared to existing lightweight YOLO-based detectors.
• This approach contributes to more effective intelligent surveillance and traffic monitoring systems.