Periodic Reporting for period 1 - AXOLOTL (From sky to seafloor observation: Achieving eXcellence in Oceanic surveiLlance and cOnservation Through deep Learning)
Periodo di rendicontazione: 2024-09-01 al 2025-11-30
During the reporting period, AXOLOTL focused on implementing its core scientific and technical activities aimed at advancing research capacity in marine and maritime artificial intelligence. The work concentrated on the development, adaptation, and validation of advanced deep-learning methodologies for marine biodiversity assessment and maritime monitoring, using heterogeneous and large-scale datasets.
Key activities included the design of data pipelines supporting the ingestion, harmonisation, and preprocessing of multi-source datasets, such as Earth observation imagery and in situ marine data. Emphasis was placed on dataset quality, annotation strategies, and reproducibility, including the definition of common protocols and standards shared among partners. Expert-driven annotation campaigns supported supervised learning approaches, addressing challenges related to class imbalance and spatial variability.
On the modelling side, AXOLOTL implemented and tested state-of-the-art deep-learning architectures, including convolutional neural networks and segmentation models tailored to complex marine environments. Experimental work explored model generalisation across geographical areas and sensor types, as well as performance optimisation under realistic data constraints, enabling systematic evaluation and iterative refinement.
Joint scientific work was complemented by staff exchanges, technical study visits, and specialised training activities, facilitating direct knowledge transfer on advanced AI workflows, model evaluation techniques, and research software and data management practices.
In parallel, AXOLOTL supported scientific coordination through a shared research roadmap aligned with European research directions, including data-driven ocean observation and digital twin approaches. Preparatory work for joint scientific publications and future research activities was initiated, ensuring continuity beyond individual tasks.
Overall, the reporting period was characterised by the execution of core scientific activities, establishment of common technical frameworks, and consolidation of collaborative research practices underpinning the project’s AI-driven marine research agenda.
Key activities included the design of data pipelines supporting the ingestion, harmonisation, and preprocessing of multi-source datasets, such as Earth observation imagery and in situ marine data. Emphasis was placed on dataset quality, annotation strategies, and reproducibility, including the definition of common protocols and standards shared among partners. Expert-driven annotation campaigns supported supervised learning approaches, addressing challenges related to class imbalance and spatial variability.
On the modelling side, AXOLOTL implemented and tested state-of-the-art deep-learning architectures, including convolutional neural networks and segmentation models tailored to complex marine environments. Experimental work explored model generalisation across geographical areas and sensor types, as well as performance optimisation under realistic data constraints, enabling systematic evaluation and iterative refinement.
Joint scientific work was complemented by staff exchanges, technical study visits, and specialised training activities, facilitating direct knowledge transfer on advanced AI workflows, model evaluation techniques, and research software and data management practices.
In parallel, AXOLOTL supported scientific coordination through a shared research roadmap aligned with European research directions, including data-driven ocean observation and digital twin approaches. Preparatory work for joint scientific publications and future research activities was initiated, ensuring continuity beyond individual tasks.
Overall, the reporting period was characterised by the execution of core scientific activities, establishment of common technical frameworks, and consolidation of collaborative research practices underpinning the project’s AI-driven marine research agenda.
During the reporting period, AXOLOTL focused on implementing its core scientific and technical activities aimed at advancing research capacity in marine and maritime artificial intelligence. The work concentrated on the development, adaptation, and validation of advanced deep-learning methodologies for marine biodiversity assessment and maritime monitoring, using heterogeneous and large-scale datasets.
Key activities included the design of data pipelines supporting the ingestion, harmonisation, and preprocessing of multi-source datasets, such as Earth observation imagery and in situ marine data. Emphasis was placed on dataset quality, annotation strategies, and reproducibility, including the definition of common protocols and standards shared among partners. Expert-driven annotation campaigns supported supervised learning approaches, addressing challenges related to class imbalance and spatial variability.
On the modelling side, AXOLOTL implemented and tested state-of-the-art deep-learning architectures, including convolutional neural networks and segmentation models tailored to complex marine environments. Experimental work explored model generalisation across geographical areas and sensor types, as well as performance optimisation under realistic data constraints, enabling systematic evaluation and iterative refinement.
Joint scientific work was complemented by staff exchanges, technical study visits, and specialised training activities, facilitating direct knowledge transfer on advanced AI workflows, model evaluation techniques, and research software and data management practices.
In parallel, AXOLOTL supported scientific coordination through a shared research roadmap aligned with European research directions, including data-driven ocean observation and digital twin approaches. Preparatory work for joint scientific publications and future research activities was initiated, ensuring continuity beyond individual tasks.
Overall, the reporting period was characterised by the execution of core scientific activities, establishment of common technical frameworks, and consolidation of collaborative research practices underpinning the project’s AI-driven marine research agenda.
Key activities included the design of data pipelines supporting the ingestion, harmonisation, and preprocessing of multi-source datasets, such as Earth observation imagery and in situ marine data. Emphasis was placed on dataset quality, annotation strategies, and reproducibility, including the definition of common protocols and standards shared among partners. Expert-driven annotation campaigns supported supervised learning approaches, addressing challenges related to class imbalance and spatial variability.
On the modelling side, AXOLOTL implemented and tested state-of-the-art deep-learning architectures, including convolutional neural networks and segmentation models tailored to complex marine environments. Experimental work explored model generalisation across geographical areas and sensor types, as well as performance optimisation under realistic data constraints, enabling systematic evaluation and iterative refinement.
Joint scientific work was complemented by staff exchanges, technical study visits, and specialised training activities, facilitating direct knowledge transfer on advanced AI workflows, model evaluation techniques, and research software and data management practices.
In parallel, AXOLOTL supported scientific coordination through a shared research roadmap aligned with European research directions, including data-driven ocean observation and digital twin approaches. Preparatory work for joint scientific publications and future research activities was initiated, ensuring continuity beyond individual tasks.
Overall, the reporting period was characterised by the execution of core scientific activities, establishment of common technical frameworks, and consolidation of collaborative research practices underpinning the project’s AI-driven marine research agenda.
The results generated by AXOLOTL have the potential to strengthen research and innovation capacity in marine and maritime artificial intelligence, particularly within a Widening-country context. The project’s scientific work on deep-learning methodologies, data pipelines, and AI-enabled marine monitoring provides a solid technical basis for future research, innovation activities, and integration into European initiatives such as data-driven ocean observation and Digital Twin of the Ocean frameworks.
At institutional level, AXOLOTL results are expected to enhance the long-term ability of the coordinating institution and its staff to participate in competitive European research programmes, co-develop advanced AI solutions, and contribute to interdisciplinary marine science agendas. At European level, the project supports efforts to reduce fragmentation in AI and ocean research by reinforcing collaboration between Widening and non-Widening institutions.
Further uptake of AXOLOTL results will depend on continued research and validation, including peer-reviewed publications and follow-up projects. Demonstration activities, sustained funding, and alignment with European data, AI, as well as continued international collaboration, will be important to maximise long-term impact and reuse.
At institutional level, AXOLOTL results are expected to enhance the long-term ability of the coordinating institution and its staff to participate in competitive European research programmes, co-develop advanced AI solutions, and contribute to interdisciplinary marine science agendas. At European level, the project supports efforts to reduce fragmentation in AI and ocean research by reinforcing collaboration between Widening and non-Widening institutions.
Further uptake of AXOLOTL results will depend on continued research and validation, including peer-reviewed publications and follow-up projects. Demonstration activities, sustained funding, and alignment with European data, AI, as well as continued international collaboration, will be important to maximise long-term impact and reuse.