Publication
Best practices for AI-based image analysis applications in aquatic sciences: The iMagine case study
Elnaz Azmi; Khadijeh Alibabaei; Valentin Kozlov; Tjerk Krijger; Gabriele Accarino; Igor Ruiz Atake; Sakina-Dorothée Ayata; Amanda Calatrava; Marco Mariano De Carlo; Wout Decrop; Donatello Elia; Sandro Luigi Fiore; Marco Francescangeli; Jesús Soriano-González; Jean-Olivier Irisson; Martin Laviale; Rune Lagaisse; Antoine Lebeaud; Carolin Leluschko; Germán Moltó; Antonio Augusto Sepp Neves; Enoc Martinez; Damian Smyth; Muhammad Arabi Tayyab; Vanessa Tosello; Alvaro Lopez Garcia; Dick Schaap; Gergely Sipos
In: Ecological Informatics, Vol. 90, Pages 1-37, Elsevier, Amsterdam, 7/2025.
Abstract
The iMagine project is an EU-funded initiative led by the EGI Foundation. One of the objectives of this project is to provide an AI platform that leverages AI-powered tools to improve the processing and analysis of imaging data from marine and freshwater ecosystems, contributing to the study of the health of oceans, seas, coasts, and inland waters. Connected to the European Open Science Cloud (EOSC), iMagine supports the development, training, and deployment of AI models by collaborating with twelve use cases across diverse aquatic science fields. This collaboration fosters valuable insights and accelerates scientific progress by refining existing solutions in data acquisition, preprocessing, and model deployment. The platform offers trained models as a service, integrating AI tools for image annotation, ensuring the creation of high-quality datasets that comply with FAIR principles. Through these methodologies, iMagine enhances consistency, enabling researchers to efficiently publish and share data in repositories.
Beyond its AI tools, iMagine places a strong emphasis on deep learning models, such as convolutional neural networks, for tasks like image classification, object detection, and segmentation, tailored to the unique requirements of aquatic sciences. It also provides robust performance evaluation tools, including experiment tracking, while tackling challenges such as AI model drift and data biases to ensure research reproducibility and transparency. The platform enables users to develop, train, share, and deploy AI models within a flexible environment that integrates with federated cloud and high-performance computing infrastructures, using Docker containers for smooth execution. Additionally, iMagine fosters collaboration with projects like AI4EOSC and Blue-Cloud, and Research Infrastructures such as EMSO and SeaDataNet, expanding its impact on the scientific community.
This paper summarizes the key lessons and best practices learned in the iMagine project through the full process of AI-based aquatic image analysis, from data preparation and annotation to model deployment and evaluation. The paper therefore helps aquatic scientists advance their AI-driven image analysis approaches.