Automatic detection of weld lines on underwater hull surface using unpaired image translation

Weld lines typically found on the hull surface of underwater structures, including ship hulls, can provide useful information for estimating the position of underwater hull-cleaning robots. Although recent deep learning-based object detection methods can be employed to automatically detect weld line features in optical images taken by hull-cleaning robots, the performances of a weld line detection model trained with one dataset may be significantly degraded in visually different weld line images obtained from hull surfaces never visited before. To overcome this limitation, this study presents an automatic weld line detection framework by integrating a deep learning-based object detection model and an unpaired image translation method. Experimental results are shown to demonstrate the practical feasibility of the proposed method using underwater image datasets obtained in field experiments with full-scale ships in real sea environments.