Identification of manganese crusts in 3D visual reconstructions to filter geo-registered acoustic sub-surface measurements

The volumetric distribution of cobalt-rich manganese crusts (CRC) is of significant interest for mining and geology. Traditionally studying underwater deposits of CRC involved physical sampling from Remotely Operated Vehicles (ROV) or using dredges. Recently, acoustic measurements of crust thickness have been demonstrated that can give a significantly higher spatial resolution for measurement. The probe makes high resolution measurements of sub-surface reflections to calculate the thickness of the deposit. However, CRC coverage is often not continuous and it is difficult to determine from the acoustic signals alone whether an acoustic signal was measured over CRC or not. The authors propose a method to filter these using visual information (3D colour map of the seafloor) from the same region. After locating the acoustic beam on the seafloor, points around the region are selected. A number of analyses are performed on this point cloud to extract parameters that can reliably discriminate between exposed CRC and other types of seafloor. The proposed method was tested on two areas of seafloor - one which is known to contain crust and one which does not contain crust. These two sets of data were then used to train a Support Vector Machine (SVM) classifier. The trained classifier was then tested with the training sets and a test set containing both crust and non-crust regions to verify if the CRC is being detected reliably. The results were promising; in 85.4% of the cases, the detection was successful. The performance will be verified by using larger sets of data. In our future work, the results can be applied to estimate a volumetric distribution of CRC in the region.