Polar View is supporting the Norwegian Computing Center in their project for ESA to develop, train, and implement a multi-modal foundation model (FM) and use the FM as the basis for the development and implementation of six different use-cases. Polar View is providing the iceberg detection use case.
Foundation models provide a paradigm shift in constructing AI models. They are trained using self-supervised learning on a large amount of unlabelled data, and by learning at scale on this vast quantity of data, FMs can capture complex patterns in the data. The FMs are pre-trained models that can be adapted to a so-called downstream task via finetuning or few-shot/zero-shot learning strategies and can serve as a basis to create powerful applications on a wide range of domains.
The goal of the iceberg use case is to improve the capability to detect and monitor icebergs using satellites. FMs will help address the three major challenges facing this task:
- Identifying icebergs in sea ice
- Differentiating icebergs from ships
- Identifying small icebergs, especially in rough conditions