SEIDOR – ARIM – Autonomous Robotic Sea-Floor Infrastructure
SEIDOR
Sector: Industry
Business Case
Within the scope of a European project, Deusto SEIDOR is responsible for developing artificial vision software to detect, classify, and automatically count benthopelagic species (those living near the bottom) from images captured by aquatic observatories (LoVe Observatory), and autonomous aquatic robots (crawlers).
Objectives
The project’s primary goal is to create a robotic infrastructure for monitoring and automatically analysing the seabed. This infrastructure aims to assess the biodiversity of the seabed in a specified area, enabling the evaluation of ecological impacts prior to conducting oil and gas activities on the seabed.
Use case
This image analysis is currently carried out manually by experts in marine biology, viewing the videos and noting down the species identified. To avoid this tedious work, a service was developed that indicates the location of the videos and conducts the analysis and counting of detected species, returning a report of results.
Infrastructure
Cloud
Technology
Computer vision Machine learning and deep learning
Data
Lofoten-Vesterålen Ocean Observatory (LoVe Observatory) in the Norwegian Sea (https://loveocean.no/about-love) Seafloor dataset collected by the Rossia 3000 crawler off the coast of Canada (property of ICM-CSIC)
Resources
Research personnel from Deusto SEIDOR for the definition of Machine Vision models and research personnel from the Institute of Marine Sciences (ICM-CSIC) for the definition and labelling of the datasets. SEIDOR’s high performance CPD infrastructure for training and inference service in SEIDOR’s CPD.
Difficulties and learning
Challenges were faced in obtaining the crawler dataset and collaborating with ICM-CSIC researchers. To address this, a dedicated tool was developed to enable remote labelling by various researchers.
KPIs (business impact and metrics of the model)
Confusion matrix AUROC Loss + accuracy
Funding
Financed with European funds (Era-Net) through the CDT.
Collaborators, Partners
Institute of Marine Research (Norway), Metas AS (Norway), GEOMAR Helmholtz Centre for Ocean Research (Germany), iSeaMC (Germany), Kraken Robotik (Germany), ICM-CSIC (Spain).
