The current process for extracting information from zooplankton images involves separate applications, like UVPapp for data processing, EcoPart for particle information, and EcoTaxa for image sorting with machine learning assistance. These applications are connected through manual export-import operations, and a manual validation step is required for the final taxonomic classification of each image.
To handle large datasets that can't be manually verified, ANERIS will develop AIES-ZOO, which will include: (i) building APIs to automate communication between applications and improving them when possible, (ii) training AI classifiers with high taxonomic resolution for image sorting and operational deployment, (iii) establishing a periodic validation process for a subset of data to control for errors due to dataset shifts, and (iv) creating routines to prepare both automated and manually validated datasets in formats accepted by major biodiversity databases.
Currently, EcoPart is being rewritten to allow application for UVP data, including both particle and plankton data which would then be loaded and annotated in Ecotaxa for taxonomic identification. Main focus is on optimizing the backend to manage users for the application.
Governmental Authorities: Fshery and aquaculture institutions
Academia: Marine Research Observatories