AI-based Detection and Classification of Seafloor Litter

Evaluate the use of artificial intelligence methods for the detection and classification of seafloor litter in deep-sea imagery retrieved from the Ifremer video archive and complemented with high-resolution mosaics produced during the 2023 and 2024 MOMARSAT cruises campaigns. 

Deliver an AI-ready annotated dataset and test preliminary detection models (YOLOv8).

• Conduct a proof of concept (PoC) to assess the feasibility of applying AI methods for detecting and classifying seafloor litter in deep-sea imagery.
• Develop and deliver an AI-ready annotated dataset of seafloor images.
• Test preliminary detection models using YOLOv8 to establish baseline performance.
• Evaluate different training strategies (from scratch, COCO-pretrained, UNO-pretrained) and data-splitting approaches to optimise detection results.
• Contribute to future harmonisation efforts with European partners and explore citizen science integration (e.g., Ocean Spy) for annotation.

• Demonstrated the technical feasibility of using AI (YOLOv8) for seafloor litter detection.
• Created and made available a publicly accessible annotated dataset.
• Provided a foundation for future dataset expansion, cross-cruise integration, and collaboration with citizen science and European partners.
• Contributed a proof-of-concept baseline that can guide future improvements in marine litter detection using AI.