A lightweight underwater biological object detector with enhanced cross-scale feature interaction
- 1Yazhou Bay Innovation Institute, Hainan Tropical Ocean University
- 2College of Marine Science and Technology, Hainan Tropical Ocean University
Abstract
Underwater biological object detection is important for intelligent marine monitoring, yet its performance is often limited by severe image degradation, background clutter, and large variations in target scale. These factors can weaken feature representation during multi-scale fusion and reduce localization reliability in lightweight detectors. In this study, we propose UD-YOLO, a lightweight underwater object detector designed to improve cross-scale feature interaction and localization stability under degraded underwater conditions. The framework enhances contextual representation in the backbone, stabilizes bidirectional information propagation in the neck, and improves multi-scale localization with a lightweight shared detection head and scale-aware regression supervision. On the RUOD dataset, UD-YOLO improves [email protected]:0.95 by 2.1% points over YOLOv11n, while reducing parameter count by 0.1 M and computational cost by 0.2 GFLOPs. Additional evaluations on the URPC and DUO show consistent gains over YOLOv11n under different underwater benchmark settings. These results suggest that UD-YOLO provides an effective accuracy-efficiency trade-off for lightweight underwater biological object detection.
Related
- LDA-YOLO11: a lightweight detail-attention detector for edge-deployable road surface defect detection
- Enhanced YOLOv8s for lightweight asphalt pavement crack detection under complex road conditions
- Integrating decision trees with ANOVA for enhanced interaction visualisations
- Pott's puffy tumour in a healthy adolescent following sinus barotrauma from underwater training.
- Underwater hybrid ESD versus conventional ESD for colorectal laterally spreading tumors: a retrospective study
BibTeX
@article{zhu2026lightweight,
title = {{A lightweight underwater biological object detector with enhanced cross-scale feature interaction}},
author = {Xiaolong Zhu and Jiayu Wang and Yukang Wang and Xiaoju Pan and Haitao Guo and Xiangzi Chen},
journal = {Scientific Reports},
year = {2026},
month = jul,
doi = {10.1038/s41598-026-63955-5},
url = {https://doi.org/10.1038/s41598-026-63955-5},
note = {CC BY},
}
RIS
TY - JOUR
TI - A lightweight underwater biological object detector with enhanced cross-scale feature interaction
AU - Xiaolong Zhu
AU - Jiayu Wang
AU - Yukang Wang
AU - Xiaoju Pan
AU - Haitao Guo
AU - Xiangzi Chen
PY - 2026
DA - 2026/07/01
JO - Scientific Reports
DO - 10.1038/s41598-026-63955-5
UR - https://doi.org/10.1038/s41598-026-63955-5
AB - Underwater biological object detection is important for intelligent marine monitoring, yet its performance is often limited by severe image degradation, background clutter, and large variations in target scale. These factors can weaken feature representation during multi-scale fusion and reduce localization reliability in lightweight detectors. In this study, we propose UD-YOLO, a lightweight underwater object detector designed to improve cross-scale feature interaction and localization stability under degraded underwater conditions. The framework enhances contextual representation in the backbone, stabilizes bidirectional information propagation in the neck, and improves multi-scale localization with a lightweight shared detection head and scale-aware regression supervision. On the RUOD dataset, UD-YOLO improves [email protected]:0.95 by 2.1% points over YOLOv11n, while reducing parameter count by 0.1 M and computational cost by 0.2 GFLOPs. Additional evaluations on the URPC and DUO show consistent gains over YOLOv11n under different underwater benchmark settings. These results suggest that UD-YOLO provides an effective accuracy-efficiency trade-off for lightweight underwater biological object detection.
SN - 2045-2322
ER -