Description
Machine learning applications require a large amount of harmonized training data to be effective. In the marine sciences, a wealth of data already exists, but the data often lacks sufficient metadata to harmonize different datasets. FAIR (Findable, Accessible, Interoperable, Reusable) digital objects can bridge the gap between different datasets by providing structured and standardized metadata, which is essential for harmonizing large numbers of datasets.
This FAIRness can be achieved for photos and videos with iFDOs - FAIR Digital Objects for images.
FAIRness starts directly after image capture by describing the data with FAIR metadata. It includes core information (e.g., ownership, autorized use, geospatial information), capture information (e.g., deployment and camera position) and content information (e.g., annotations and particle count) - information that is structured, standardized and term-based. Having such a human- and machine-readable data format makes it easier to interact with the data, i.e., to implement generic applications that can handle the data.
We are developing tools and services to lower the barrier to entry for creating and publishing iFDOs: The iFDO Creator is a GUI desktop and web application that utilizes existing metadata repositories (e.g., O2A Registry) to enable manual, but tool-assisted, creation and editing of iFDOs. The planned services, O2A Ingest and O2A Workspaces, will provide access to storage and computing environments, enabling researchers to work on ever larger datasets without having to provide the hardware themselves. We are also contributing to and utilizing the Marimba framework to automate image dataset processing and iFDO creation.
These FAIR datasets are then published to the scientific data repository PANGAEA and ingested into our spatial data infrastructure, allowing them to become findable and accessible with web GIS (Geographic Information System) applications, i.e., Marine Data.
Our efforts aim to increase the adoption of iFDOs in the marine imaging community, not only benefiting the impact of researchers but also the interoperability between applications and datasets.
