RGU Computer Vision for the Energy Sector Industry Project Course

A challenging 12 week SQF11 Level MSc real life industry project focussing on image classification of underwater images to identify corroded subsea items/structures. Skills learnt included image enhancement techniques, CNN transfer learning, tensorflow.

I collated over 5000 subsea images, the images were enhanced using underwater image enhancement techniques and trained an Inception based CNN to classify underwater images either as a corroded item or non corroded. The final model accuracy was 92%.

This project aligned with my career as a Subsea controls Engineer where I often review underwater ROV/Diver videos and appreciate the visibility issues dealing with such images.

This is the starting point in a development to utilise ML techniques to support automated subsea monitoring to identify structures, structure types and anomalies. Such monitoring is a mandatory requirement of all operators.