Deep Learning for Automated Corrosion Detection

Corrosion costs the economy an estimated 3% of GDP annually. For asset owners’, visual inspection is still the primary means for discovering and rectifying issues arising from corrosion. This project was conceived to explore the current state of the art for deep learning computer vision that is so successful with cats, dogs and faces; and test its performance when faced with “rust”. The thesis presents the history of automating corrosion detection, outlines the metrics and frameworks for success, and identifies the barriers required to be overcome. The research outcome culminated in deep learning detection models that approach human accuracy.