Assessing benthic marine habitats colonized with posidonia oceanica using autonomous marine robots and deep learning: A Eurofleets campaign

Abstract

This paper describes a proof-of-concept of new processes aimed to observe and analyze marine ecosystems, overcoming common bottlenecks, namely, data gathering and their precise geo-localization, data classification, scene recognition, limited extension of missions, and restricted depths. Field operations were carried out with three different autonomous platforms, an Autonomous Underwater Vehicle (AUV), an Autonomous Surface Vehicle (ASV) and a Lagrangian Drifter (LD), all deployed and recovered from the RV SOCIB support vessel, during a seven-day campaign in the Spanish marine and terrestrial protected area of Cabrera, Balearic Islands. Activities were designed and planned to collect video sequences in situ, from the AUV and from the LD, both navigating close to the seafloor in certain particular locations of the park colonized with the endemic seagrass species Posidonia oceanica (Po). Visual data was later composed in photo-mosaics and classified using machine learning algorithms. The joint application of all these technologies goes one step beyond the solutions already existing in the literature based on divers or remote sensing, and, in line with other solutions that apply underwater vehicles to monitor underwater ecosystems, it extends the range of operations for Po control in time, space and depth, obtaining the data in situ, eliminating risks to humans and automating the computation of several parameters that define the estate of the ecosystems. Results obtained from field experiments were compared to existing habitat distribution maps to infer qualitative conclusions about the evolution of the seagrass meadows. The methodology can be scaled to other locations and mission extensions, and focused on other species by adapting the image processing software, with known limitations such as data storage and battery endurance.