An edge-aware and multi-scale shore-based ship detector for small vessels under adverse environmental conditions
Shore-based ship detection is essential for coastal security and intelligent port management, yet adverse sea conditions, vessel occlusion and scale variations hinder reliable detection. This study proposes DE-YOLO, a multiscale feature-enhanced ship detector based on the YOLOX architecture. An edge-residual-guided adaptive feature enhancement module captures local structural variations and generates adaptive weights to recalibrate features, improving small-vessel detection against complex maritime backgrounds. A dynamic multiscale contextual fusion module integrates features at different scales through multiscale convolutional attention, strengthening the representation of vessels with varying sizes. Efficient intersection over union loss is adopted to supervise bounding-box regression. Experiments on the Singapore Maritime Dataset yield mean average precision values of 90.4% at an intersection over union threshold of 0.5 and 63.2% averaged over thresholds from 0.5 to 0.95. These results indicate the potential of DE-YOLO for practical shore-based maritime surveillance.
Authors
- Ruoling Deng (ORCID: https://orcid.org/0000-0002-9207-9283)
- Tujie Wu (ORCID: https://orcid.org/0009-0006-6832-971X)
- Haitao Liu (ORCID: https://orcid.org/0000-0003-0995-4655)
- Hubo Chu
- Zhou Yang
Institutions
- Guangdong Ocean University (CN)
Publication Details
- Journal
- Ships and Offshore Structures
- Published
- 2026-11-21
- DOI
- https://doi.org/10.1080/17445302.2026.2765718
- Primary Topic
- Advanced Neural Network Applications
- Type
- article
- Field-Weighted Citation Impact
- 0.00