| Inspecting circumferential welds in caissons is a critical task in the offshore industry for ensuring the safety and reliability of structures. However, identifying and classifying different types of circumferential welds can be challenging in subsea environments due to low contrast, variable illumination, and suspended particles. To address this challenge, we present a framework for automating the classification of circumferential welds using deep learning-based methods. We used a dataset of 4,000 images for experimental purposes and utilised three state-of-the-art pre-trained Convolutional Neural Network (CNN) architectures, including MobileNet V2, Xception, and EfficientNet. Our results showed superior performance of EfficientNet, with high levels of accuracy (86.75%), recall (91%), and F1-score (87.29%), as well as demonstrating efficient time. These findings suggest that leveraging deep learning-based methods can significantly reduce the time required for inspection tasks. This work opens a new research direction toward digitally transforming inspection tasks in the Oil and Gas industry. |
*** Title, author list and abstract as seen in the Camera-Ready version of the paper that was provided to Conference Committee. Small changes that may have occurred during processing by Springer may not appear in this window.