Zhang, Mingxin, Li, Xiaolei, Sun, Guangbo, Zhang, Youmei, Rosin, Paul L.  ORCID: https://orcid.org/0000-0002-4965-3884 and Zhang, Wei
2025.
SPG-GT: structural prior guided GNN-transformers for ship landmark detection.
IEEE Journal of Oceanic Engineering
10.1109/joe.2025.3617916
|
Abstract
The visual perception of ships has gained increasing attention in computer vision and ocean engineering. Ship landmark detection plays a crucial role in various applications, including ship detection, ship recognition, ship image generation, and ship area detection. However, existing methods for landmark detection did not fully leverage the association among landmarks, which leads to a lack of overall perception of ships and limits the performance of ship landmark detection. To address this issue, this article proposes SPG-GT, a ship landmark detection model that combines graph neural networks (GNNs) and transformers guided by ship structural prior. GNNs effectively encode the connectivity information between ship landmarks, a set of keypoints important for defining the overall structure of a ship. SPG-GT also leverages transformers and coordinate convolution to extract global and local features of a ship, which ensures that the detected landmarks are consistent with the nature of a ship. We evaluate SPG-GT on the publicly available SLAD data set and a newly created SLAD++ data set. The experimental results on both data sets demonstrate the superior performance of SPG-GT.
| Item Type: | Article |
|---|---|
| Date Type: | Published Online |
| Status: | In Press |
| Schools: | Schools > Computer Science & Informatics |
| Additional Information: | License information from Publisher: LICENSE 1: URL: https://ieeexplore.ieee.org/Xplorehelp/downloads/license-information/IEEE.html, Start Date: 2025-01-01 |
| Publisher: | Institute of Electrical and Electronics Engineers |
| ISSN: | 0364-9059 |
| Date of Acceptance: | 30 August 2025 |
| Last Modified: | 20 Nov 2025 14:45 |
| URI: | https://orca.cardiff.ac.uk/id/eprint/182544 |
Actions (repository staff only)
 |
Edit Item |
Dimensions
Dimensions