Accurate biological detection is crucial for autonomous navigation of underwater robots, yet severely challenged by optical degradation and scale variation in marine environments. While image enhancement and domain adaptation methods offer some mitigation, they often operate as disjointed preprocessing steps, potentially introducing artifacts and compromising downstream detection performance. Furthermore, existing architectures struggle to balance accuracy, computational efficiency, and robustness across the extreme scale variability of marine organisms in challenging underwater conditions. To overcome these limitations, we propose UW-YOLO-Bio, a novel framework built upon the YOLOv8 architecture. Our approach integrates three dedicated modules: (1) The Global Context 3D Perception Module (GCPM), which captures long-range dependencies to mitigate occlusion and noise without the quadratic cost of self-attention; (2) The Channel-Aggregation Efficient Downsampling Block (CAEDB), which preserves critical information from low-contrast targets during spatial reduction; (3) The Regional Context Feature Pyramid Network (RCFPN), which optimizes multi-scale fusion with contextual awareness for small marine organisms. Extensive evaluations on DUO, RUOD, and URPC datasets demonstrate state-of-the-art performance, achieving an average improvement in mAP50 of up to 2.0% across benchmarks while simultaneously reducing model parameters by 8.3%. Notably, it maintains a real-time inference speed of 61.8 FPS, rendering it highly suitable for deployment on resource-constrained autonomous underwater vehicles (AUVs).

Underwater detection is crucial for the autonomous operation of Autonomous Underwater Vehicles (AUVs). However, underwater environments pose significant challenges, including severe image degradation, complex target deformation, and densely distributed small objects. Most existing methods treat image enhancement as an independent preprocessing module and rely on fixed-shape convolution kernels for feature extraction, which often leads to inconsistent optimization objectives and limited capability in handling irregular targets and fine-grained small-object details. To address these issues, we propose an End-to-End Adaptive Underwater Detection framework (E2E-AUD). Specifically, a lightweight image enhancement module, UnitModule, is embedded into the detection network so that enhancement can be jointly optimized with detection and directly serve downstream feature learning. In addition, linear deformable convolution (LDConv) is introduced into the backbone to adaptively model polymorphic targets, while Haar wavelet downsampling (HWD) is adopted to preserve boundary and texture information through frequency-domain analysis. Experiments on the DUO and URPC datasets demonstrate that E2E-AUD achieves superior performance over both general-purpose and underwater-specific detectors. Specifically, on the DUO dataset, our model reaches 86.2% mAP50 and 67.8% mAP50-95, outperforming the recent YOLOv12 by 3.0% and 2.7%, respectively. On the highly turbid URPC dataset, it achieves 84.3% mAP50 and 50.8% mAP50-95, surpassing the competitive underwater-specific detector LEFEN by notable margins in strict localization metrics. Furthermore, E2E-AUD maintains a real-time inference speed of 21.8 FPS with highly constrained computational complexity (9.4 GFLOPs), proving its exceptional balance between detection accuracy and deployment efficiency compared to previous methods.