Efficient UAV-Based Maritime Search and Rescue: A Multi-Scale Edge-Enhanced Detection Algorithm with Frequency-Spatial Collaborative Processing

Unmanned aerial vehicle (UAV) detection of maritime targets faces significant challenges including dense small target distribution, complex sea surface backgrounds, and severe wave interference. This paper proposes FM-METR, an efficient detection algorithm addressing these issues through three key innovations. First, it reconstructs the backbone network using a spatial-semantic information coordination mechanism with feature complementary mapping and multi-core perception strategies to enhance small target feature extraction while reducing computational complexity. Second, a multi-scale edge enhancement architecture achieves precise localization of critical targets through adaptive parallel processing and dual-domain intelligent fusion. Third, a frequency domain adaptive processing mechanism effectively separates target signals from sea surface noise via multi-scale frequency domain decomposition. Experimental results demonstrate superior performance: on the Self-SeaDrone dataset, FM-METR achieves 92.9% mAP@0.5, improving 3.3 percentage points over baseline; on SeaDroneSea dataset, it attains 86.6% mAP@0.5 and 93.2% detection accuracy, with improvements of 2.9 and 2.7 percentage points respectively. The model exhibits remarkable efficiency with only 19.4 MB size and 28.1 GFLOPs computational complexity, achieving 32.9% and 51.8% reductions compared to baseline. Real-time processing capabilities are demonstrated with 88.3 FPS on PC platforms and 47.8 FPS on RK3588 embedded devices, fully satisfying maritime search and rescue requirements. The proposed method shows superior performance and practical application value under complex sea conditions.