Karthik Iyer ^1* Camilla Marnor ¹ Daniel W Schmid ¹ Ebbe H Hartz ²

• ¹ Bergwerk AS, Sandefjord, Norway
• ² Aker BP, Lysaker, Norway

We present a new dataset combined with an advanced object detection model designed to rapidly and accurately detect benthic lifeforms in deep-sea environments of the North Atlantic. When benchmarked against a published dataset from the same region, the DeepSee model demonstrates high performance metrics, achieving an impressive overall mean Average Precision (mAP) score of 0.84. Additionally, the model returns very few false positives, ensuring reliable detection. While the model is not intended to replace the expertise of experienced biologists, it significantly reduces the annotation workload and increases processing speed by over 1000 times. As more data becomes available, adding to the dataset and retraining the model will result in continued enhancements in its detection capabilities. Importantly, this dataset can be expanded to include other forms of benthic life found in the North Atlantic and other regions, facilitating broader ecological studies and applications. The deployment of this model will thus enable the creation of high-resolution maps of benthic life on the largely unexplored ocean floor of the Norwegian Continental Shelf (NCS) and beyond. This capability will aid in informed decision-making regarding marine resource exploration, including mining operations and bottom trawling, as well as deep-sea pipeline laying. It will ultimately contribute to marine protection efforts and support the sustainable management of deep-sea resources and ecosystems.