ULO: An Underwater Light-Weight Object Detector for Edge Computing

Recent studies on underwater object detection have progressed with the development of deep-learning methods. Generally, model performance increase is accompanied by an in computation. However, a significant fraction remotely operated vehicles (ROVs) and autonomous (AUVs) operate environments limited power computation resources, making large models inapplicable. In this paper, we propose fast compact detector—namely, Underwater Light-weight Object detector (ULO)—for several marine products, such as scallops, starfish, echinus, holothurians. ULO achieves comparable results to YOLO-v3 less than 7% its modified based YOLO Nano architecture, some modern architectures are used optimize it, Ghost module decoupled head design detection. We adaptive pre-processing for image degradation problem that common images. The lightweight simple use, ablation experiments verify effectiveness. Moreover, Tiny, lite version ULO, proposed achieve further reduction. Furthermore, annotations URPC2019 dataset, more accurate localization classification. refined available public research use.