Results validate that MSFE-YOLO achieves an effective balance between accuracy and efficiency for underwater object detection, with notable advantages in detecting small-scale targets.
Abstract
Underwater target detection has become a prominent research area in computer vision, while high noise levels, low visibility, and blurred edges pose significant challenges, which severely degrade the performance of existing detection algorithms. Such challenges necessitate detection models that simultaneously achieve high accuracy, operational stability, superior generalization, real-time inference, and architectural efficiency. To address these challenges, we propose the multi-scale feature enhancement YOLO (MSFE-YOLO), which integrates four key improvements: First, the CSP-PMSFA module is proposed to preserve multi-scale information and enhance feature fusion. Second, the C2PSA-EDFFN module is designed to effectively capture frequency-domain information and provide complementary local details. Third, the DetectAux module is incorporated to address feature learning inconsistencies in multi-scale detection through deep supervision and hierarchical label assignment. Finally, the improved D-InterpIoU loss is introduced as a novel objective function that benefits the training stages. On the URPC dataset, MSFE-YOLO achieves mAP50 and mAP50:95 of 86.0% and 50.9%, surpassing YOLOv11n by 1.7% and 1.4%, respectively. On the UTDAC dataset, MSFE-YOLO achieves mAP50 and mAP50:95 of 84.3% and 50.2%, exceeding YOLOv11n by 1.2% and 0.9%, respectively, while maintaining a real-time inference speed of over 300 FPS on an NVIDIA RTX 3090 Ti. These results validate that MSFE-YOLO achieves an effective balance between accuracy and efficiency for underwater object detection, with notable advantages in detecting small-scale targets.
A novel underwater target detection framework that integrates feature enhancement with semantic-spatial guided fusion, built upon the RT-DETR architecture, that significantly reduces false positives and missed detections while maintaining real-time performance is proposed.
P. Parashar, A. Kushwah· Discover Computing· 0 citations
Experimental results demonstrate that the proposed Bidirectional weighted Concat with Efficient multi-scale Attention You Only Look Once (BCEA-YOLO) method outperforms competing methods for underwater object detection, and edge deployment tests on the NVIDIA Jetson platform validate its real-time inference efficiency.
Xing-Yu Wang, Yu-Han Lin, Quan J. Wang et al.· Intelligent Marine Technolog...· 0 citations
To address the inherent limitations of the YOLOv12n model in complex underwater environments, such as insufficient global context capture, restricted detection accuracy, and feature information loss during downsampling, this paper proposes a lightweight enhanced model named WDF-YOLO. By analyzing the baseline model’s c...
Tien-Wen Sung, Yi-Xin Chen, Chao-Yang Lee et al.· Journal of King Saud Univers...· 0 citations
Underwater object detection is challenged by visual degradation such as low contrast, background noise and the difficulty of identifying small targets. To alleviate these issues, we propose UFL-YOLO, an enhanced version of YOLOv10 augmented with lightweight yet effective modules: small object enhance pyramid (SOEP), un...
Yu-Xi Wu, Yan Wang, Jing Zhang et al.· Engineering Research Express· 0 citations
Real-time object detection under low illumination remains challenging because weak contrast, sensor noise, and feature degradation can reduce target discriminability, while practical deployment also requires careful control of model complexity and computational cost. Existing enhancement-then-detect pipelines can impro...
Min Hao, Zhe Chen, Ji-Long Li et al.· IEEE Access· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.