Automatic benthic habitat mapping using inexpensive underwater drones

Abstract

The generation of benthic habitat maps relies either on direct in-situ observations made by SCUBA divers swimming in a rectilinear fashion, or on costly remote sensing techniques involving either ROVs or sonar technology. The recent commercialisation of off-the-shelf underwater drones has enhanced benthic mapping possibilities by providing a costeffective alternative. Despite still requiring ground truthing, such drones do not rely extensively on boat-support services. In this study, the applicability and feasibility of using an underwater high-resolution optical platform to automate the generation of benthic maps is investigated. A PowerVision PowerRay [1] was used to capture underwater imagery in an embayment along the north-east coast of the island of Malta (central Mediterranean). A Machine Learning method based on Self-Organizing Maps was then implemented to automate the classification process. Results produced from this technique were evaluated in terms of their accuracy through comparisons with a benthic habitat map of the same area that was generated through conventional means in a previous study.