Automated detection of sacrificial anodes on offshore wind farms for ROV inspections

In this research, two transfer learning models were trained in order to detect sacrificial anodes on the base of a floating offshore wind turbine with the aim to aid the ROV pilot in navigation. Two models were trained on a dataset of sacrificial anodes collected by our team at WindFloat Atlantic windfarm in collaboration with OceanWinds. This data was then labelled and one model was trained and validated to detect the anodes. This workflow was then further tested in a tank with anodes manufactured from 3D printed materials to test the generalisability of the workflow using the second transfer learning model. The models performed well, one achieving recalls of 85% and the other demonstrating robustness by detecting anodes in the tank despite their vastly different appearance and absence from the training data.