Veiga, RicardoRodrigues, Joao2026-04-072026-04-072024-11-13979-8-4007-0729-2http://hdl.handle.net/10400.1/28604An inclusive society actively seeks the equitable and respectful participation of all its members, regardless of their differences. This concept goes beyond tolerating diversity; it involves valuing and respecting each individual, ensuring everyone has equal access to information and opportunities, and actively participating in all aspects of life. To have a full inclusive society, we have to measure and monitor our impact in the environment, specifically in the oceans and marine life. This paper addresses this challenge by proposing a framework that leverages data aggregation and advanced machine learning techniques for Fine-Grained Visual Classification of marine species. Our methodology employs the Swin Transformer architecture, enhanced with the Fine-Grained Visual Classification Plug-in Module, to process and classify diverse marine datasets. We aggregated multiple marine datasets, preprocessed them to eliminate invalid entries, and trained our model on the refined dataset. Our findings demonstrate that dataset aggregation significantly enhances model accuracy and robustness, especially for large-scale models. Notably, the aggregated data model achieved 94.75% overall accuracy on a dataset comprising 2,548 classes and 391,374 images, compared to 85.93% on individual datasets like WildFish++.engComputer visionData aggregationData fusionFine-grained visual classificationSwin transformerjournal article10.1145/3696593.3696637