Convolutional Neural Network Approach to Detect Underwater Pipeline Degradation using IOT dataset

Abstract

In the offshore petroleum and natural gas industry, corrosion of subsea pipelines is thought to be a significant problem. It has a direct impact on the pipeline's integrity, which in turn results in gaps and leaks. Subsea visual examination and surveillance are carried out by trained human divers at the present time; moreover, offshore facilities are migrating from shallow seas to deep oceans as a result of the depletion of fossil fuels. Accordingly, for visual surveillance and inspection of subsea pipelines, an imaging-based robotic solution is required as an alternative due to the inhospitable underwater environmental factors that human divers must operate in. Absorption and light scattering which further results to blurring, colour attenuation, and low contrast, is a challenging issue for underwater imaging-based surveillance and inspection operations. This problem is caused by an unfriendly medium. As a result, a system that has been proposed could make it possible for an unmanned underwater vehicle to identify damaged pipelines in a series of images. Transfer learning from Convolutional Neural Networks (CNN) that had been previously trained was the foundation for the classifiers (CNN). Because of this, it is possible to achieve favourable results in spite of the limited number of damaging training scenarios which are faced. The method that has been suggested has been put through its paces by utilising IOT data taken from a real pipeline inspection. When estimating the amount of corrosion, a reasonable level of accuracy was achieved, which assisted in differentiating between the corrosion and non-corroded surface of corroded pipelines? The both qualitative and quantitative studies both show encouraging results, which motivate the integration of the proposed technique into a robotic system that is capable of performing underwater pipeline corrosion investigation in real time.