Automatic detection of ship draft marks and draft reading are critical tasks in intelligent port management and maritime monitoring. This paper presents a lightweight ship draft detection method based on DeepLabv3+, incorporating the SeaFormer backbone network, DenseASPP module, and PointRend boundary refinement technique. The proposed method achieves efficient waterline segmentation and accurate draft depth estimation. The SeaFormer backbone significantly enhances feature extraction capability while reducing computational complexity through its squeeze-enhanced axial attention mechanism. The DenseASPP module strengthens the perception of waterline regions by capturing multi-scale contextual information, and the PointRend module further refines boundary details. Experimental results demonstrate that the proposed method outperforms mainstream segmentation models in terms of segmentation accuracy (mIoU), parameter efficiency, and real-time performance, validating its suitability for resource-constrained scenarios. This approach provides an efficient and reliable solution for ship draft detection and smart shipping applications.
Lightweight ship draft reading framework based on DeepLabv3+ and SeaFormer
Fourth International Conference on Electronics Technology and Artificial Intelligence (ETAI 2025) ; 2025 ; China, China
Proc. SPIE ; 13692 ; 1369259
2025-07-24
Conference paper
Electronic Resource
English