Underwater object detection (UOD) plays a pivotal role in marine robotics, ecological monitoring, and environmental protection. However, existing deep learning-based detection methods face significant challenges in underwater scenarios, including coupled frequency-spatial distortions and insufficient integration of multi-scale object features caused by optical degradation. To address these limitations, we propose UHF-UOD, a novel underwater detection algorithm. The proposed framework incorporates three fundamental innovation modules: (1) Underwater Multi-Scale Feature Extraction (UMFE) module, which synergistically combines parallel deep convolutions with multi-scale kernels alongside spatial adaptive attention mechanisms, effectively expanding the receptive field while enhancing scale-adaptive feature extraction capabilities under varying degradation characteristics; (2) Frequency-Spatial Transformer Block (FSTB) integrates Fourier Transform-based Frequency Self-Attention (FSA), Multi-Scale Spatial Self-Attention (MSSA) and Convolutional Gated Linear Unit (CGLU) to address coupled frequency-spatial degradation comprehensively; (3) Hierarchical Scale-Aware Fusion Network (HSAFN), which captures long-distance dependencies through Scale-Aware Mamba (SAMamba) and is combined with the Adaptive Progressive Fusion Module (APFM) to enable cross-scale feature interaction. Experimental validation on four public datasets (UDD, UODD, DUO, RUOD) demonstrates that UHF-UOD achieves state-of-the-art performance in underwater object detection. This research offers a high-precision and high-efficiency solution for underwater object detection. The source code is available at https://github.com/EUOD/UHF-UOD . • Proposing UHF-UOD for underwater detection via dual-domain feature fusion. • Extracting scale-adaptive features under varying degradation characteristics. • Hierarchical fusion network captures long-range dependencies across scales. • Achieving state-of-the-art results on four underwater detection datasets.
Chen et al. (Sat,) studied this question.