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

Abstract

Unmanned aerial vehicle (UAV)-based object detection plays an important role in intelligent surveillance and traffic monitoring. However, small object scales, complex backgrounds, dense distributions, and occlusions make it difficult to balance detection accuracy and real-time performance. To address these challenges, this paper proposes a frequency-saliency guided multi-scale feature fusion network based on YOLO11 for UAV object detection. Specifically, a Global Frequency Feature Extraction module is designed by integrating saliency enhancement and Dual-Tree Complex Wavelet Transform to separate high-frequency target details from low-frequency background information. In addition, a lightweight multi-scale feature fusion module with soft gating and spatial attention is introduced to enhance feature representation across scales, and a Scale-Adaptive Dropout Feature Pyramid Network is employed to enhance robust multi-scale feature learning. Experimental results on the HIT-UAV dataset show that FSMF-YOLO achieves 91.6% precision, 83.5% recall, and 89.3% mAP@50 at 101 FPS, improving mAP@50 by 2.6% over YOLO11n while maintaining real-time inference. These results suggest that frequency-saliency guidance improves small-object representation in complex UAV scenes and provides a favorable accuracy-efficiency trade-off compared with lightweight YOLO-based detectors.