UOD-YOLO: A Lightweight Real-Time Model for Detecting Marine Organisms

Yijie Xi YijieJingbo Yin^*

• Shanghai Jiao Tong University School of Ocean and Civil Engineering, Shanghai, China

Underwater object detection (UOD) is extremely challenging due to poor image quality, the complexity of underwater environments, and the resource constraints of embedded devices. To address these issues, we propose UOD-YOLO, a lightweight, real-time detection framework derived from YOLOv11n. The model incorporates three key innovations: (1) a RepViT-based lightweight backbone for efficient feature representation; (2) a scale sequence feature fusion and triple feature encoding module for enhanced multi-scale information extraction; and a lightweight detection head with a channel–position attention mechanism to improve small-object recognition. Experiments using the RUOD, UTDAC2020, and DUO datasets revealed that UOD-YOLO improved mAP50 and mAP50-95 by 3.5% and 6.9%, while reducing parameters and floating-point operations by 36.3% and 19.7% compared with YOLOv11n, achieving a real-time detection speed of 279.8 frames per second. Visualization results further confirmed the model's robustness under varying illumination and depth conditions. The compact size and high accuracy of UOD-YOLO enable practical deployment in aquaculture monitoring, continuous edge-based surveillance, and multi-sensor fusion. This work provides an efficient and reliable solution for the real-time detection of marine organisms, with significant implications for intelligent aquaculture and marine ecological monitoring.