Regulatory requirements for sub-sea oil and gas operators mandates the frequent inspection of pipeline assets to ensure that their degradation and damage are maintained at acceptable levels. The inspection process is usually sub-contracted to surveyors who utilize sub-sea Remotely Operated Vehicles (ROVs), launched from a surface vessel and piloted over the pipeline. ROVs capture data from various sensors/instruments which are subsequently reviewed and interpreted by human operators, creating a log of event annotations; a slow, labor-intensive and costly process. The paper presents an automatic image annotation framework that identifies/classifies key events of interest in the video footage viz. exposure, burial, field joints, anodes, and free spans. The reported methodology utilizes transfer learning with a Deep Convolutional Neural Network (ResNet-50), fine-tuned on real-life, representative data from challenging sub-sea environments with low lighting conditions, sand agitation, sea-life and vegetation. The network outputs are configured to perform multi-label image classifications for critical events. The annotation performance varies between 95.1% and 99.7% in terms of accuracy and 90.4% and 99.4% in terms of F1-Score depending on event type. The performance results are on a per-frame basis and corroborate the potential of the algorithm to be the foundation for an intelligent decision support framework that automates the annotation process. The solution can execute annotations in real-time and is significantly more cost-effective than human-only approaches.

Utilising deep learning image classification to automatically annotate subsea pipeline video surveys can facilitate the tedious and labour-intensive process, resulting in significant time and cost savings. However, the classification of events on subsea survey videos (frame sequences) by models trained on individual frames have been proven to vary, leading to inaccuracies. The paper extends previous work on the automatic annotation of individual subsea survey frames by comparing the performance of 2D and 3D Convolutional Neural Networks (CNNs) in classifying frame sequences. The study explores the classification of burial, exposure, free span, field joint, and anode events. Sampling and regularization techniques are designed to address the challenges of an underwater inspection video dataset owing to the environment. Results show that a 2D CNN with rolling average can outperform a 3D CNN, achieving an Exact Match Ratio of 85% and F1-Score of 90%, whilst being more computationally efficient.