Performance Analysis of Subsea Pipeline Defect Classification Using Supervised Machine Learning

Subsea pipelines are pivotal to a country’s economy. These pipelines not only ensure safe and efficient transportation of gas, oil, and other refined products but also enable continuous flow of oil and gas to onshore reserves, thereby maintaining a consistent supply to meet energy demands. However, these pipelines are exposed to extreme variations in temperature and pressure which makes them susceptible to defects early in their operational life which if left untreated can potentially lead to irreversible environmental damage. This necessitates regular integrity assessment of the subsea infrastructure. Inspection of subsea pipelines involves the arduous task of reviewing inspection video footage with extreme precision. To automate the process of inspection, a comparative study of four different multi-class classification models that can classify the underwater pipeline inspection images into three distinct classes namely corroded, cracked and healthy pipes, has been evaluated. The classifiers considered for the study are EfficientNetB0, MobileNetV2, ResNet50 and VGG-16. Among the considered models, VGG-16 has shown the highest overall classification accuracy of 99.04%.