An accurate realtime underwater object segmentation using improved dual-domain YOLOv11-UOS with physics guided adaptive enhancement and attention-boosting.

The recent development and requirement in marine exploration, defense, and autonomous navigation requires accurate detection of underwater objects. However, the presence of objects in underwater images is hard to accurately identify because of poor visibility, color distortion, light dispersion, and complicated aquatic backgrounds. This research presents an advanced instance segmentation system that integrates attention-enhancing techniques and physics guided adaptive image augmentation with the latest dual domain YOLOv11-UOS architecture. The proposed method recovers color fidelity and contrast prior to segmentation by addressing wavelength-dependent attenuation and scattering through underwater image creation models. The proposed YOLOv11-UOS has a dual attention-boosting module built in to make detection even better by getting rid of noise from unimportant regions while highlighting essential characteristics. Using the publicly available custom underwater datasets the experimental results demonstrate that the proposed method does far better than baseline approach in the segmentation accuracy. The qualitative, quantitative and ablation study results reveal that the system is more resistant to partial occlusions, turbidity, and variations in lighting. Because it is precise and light, it is also good for real-time use in underwater robotics.