ABSTRACT

Polymetallic manganese nodules (PMN) occurring at 4000-6000 m depths of major ocean basins are rich in critical elements such as Co, Ni, REE etc can be the future source of raw material supply. Exploration of PMN involves exploration and sampling, using sonar and cameras. High-quality images were collected from the Indian Ocean PMN site and segmentation algorithms were used to determine nodule abundance, coverage and size. This article presents a machine-learning-based model using a Random Forest classifier to perform segmentation by utilizing labelled images and various filter responses for quantitative estimation. Results show 80 percent accuracy for the model performance using the semantic segmentation technique and the outcomes are explained in the article.

INTRODUCTION

Deep-ocean mineral deposits are attracting attention due to their resource potential and economic importance as they could significantly contribute to future raw material supply. Polymetallic nodules are one of the mineral resources found on the abyssal plains in about 4000-6000 m water depth in major ocean basins. They form two-dimensional deposits on top or within the first 10 cm of the deep-sea sediments (Kuhn et al., 2013.). Nodules of the eastern tropical Pacific and the Central Indian Ocean are of special economic interest due to their high enrichment of metals such as Ni, Cu, Co, Mo, Li, REE, and Ga (Hein et al., 2013).

The traditional methods of seabed resource assessments mainly rely on techniques such as samplers, acoustic backscatter and underwater high-definition cameras. With a sampler e.g. using box cores, we can obtain the mineral morphology and calculate mineral abundance, but this technique cannot assess minerals over a large area. It can provide misleading data in heterogeneously populated areas and might provide misleading measurements despite an actual PMN occurrence in sparsely populated areas. Multibeam data with back scatter seafloor images gives high areal coverage, but the results lack individual mineral information (Schoening et al., 2017). Optical imaging can address these problems since it gives higher quantitative resolution than hydro-acoustics and higher areal coverage than physical sampling. Deep-sea mineral images combined with image processing algorithms can provide information about mineral distribution and abundance. The individual nodule boundary needs to be delineated from all the images captured by the underwater camera to determine the abundance and size of nodules.

This content is only available via PDF.
You can access this article if you purchase or spend a download.