FewU-Det: A few-labeled underwater object detection framework with dynamic transmission estimation and distribution-aware pseudo mining

Frequent jellyfish outbreaks pose serious threats to fisheries, tourism, and coastal industries, and their occurrence is closely linked to the polyp stage. Accurate detection of polyps and other ecologically harmful organisms is therefore of both scientific and practical importance. However, polyps are small, translucent, and often blurred in underwater images, making manual annotation costly and prone to omissions. This challenge reflects a broader issue in underwater object detection, where labeled samples are limited and annotations are sparse, restricting model performance. To address this, we propose FewU-Det, a unified framework designed for underwater detection with limited annotations, where part of the data is unlabeled and the rest contains incomplete labels. The framework achieves joint optimization of physics-inspired enhancement for detection (rather than strict physical restoration), pseudo-label generation, and detector training. Compared with state-of-the-art detectors, FewU-Det achieves mAP gains of 3.22%, 2.27%, and 2.13% on the Polyp, DUO, and RUOD datasets, respectively . These findings provide an effective approach for underwater species recognition and monitoring under limited annotations, and also offer methodological support for tracing jellyfish outbreaks and guiding ecological management. The code and datasets used in this study are publicly available at https://github.com/tty114/FewU-Det . • Frequent jellyfish outbreaks linked to the polyp stage highlight the need for accurate detection under scarce and noisy annotations. • FewU-Det is proposed as a unified framework for underwater object detection with few labels. • Dynamic image restoration and pseudo-label mining enhance detection of small and blurred targets. • Joint loss design improves robustness under degraded underwater imagery. • The method provides a practical tool for underwater monitoring and supports jellyfish outbreak management.