AI against bycatch: cutting-edge technology to mitigate marine megafauna incidental capture

The incidental capture of marine species, also known as bycatch, represents a threat to the conservation of marine biodiversity. It accounts for approximately 40% of global fishing catches and particularly affects marine megafauna, including turtles, marine mammals, and sharks. In this context, two unprecedented studies led by researcher Veronica Nieves from the Universitat de València (UV) adapt and combine different AI-based methods to automatically identify incidentally captured species from images taken onboard fishing vessels. They propose various hybrid frameworks specifically tailored to these data, species, and monitoring conditions.

This research, developed at the Image Processing Laboratory (IPL) and the Cavanilles Institute of Biodiversity and Evolutionary Biology (ICBiBE)of the UV within the framework of the European Union's REDUCE project (HORIZON program), demonstrates how the integration of complementary approaches and dataset-specific tuning can offer optimized solutions. This synergy is essential to maintain species-level precision when working with low-quality images.

Veronica Nieves, founder and leader of AI4OCEANS —a research group in artificial intelligence for the oceans hosted at the UV— highlights that "AI can help transform complex images into reliable biodiversity information. By combining different models that offer a balance between speed and precision, we come closer to practical monitoring tools capable of improving bycatch estimations. These advances contribute to enhancing the protection of vulnerable marine species".

Among the explored model combinations, frameworks that combine object recognition in images with others that locate and identify the species appearing in them stand out. Examples include the integration of Convolutional Neural Networks (CNN) withRandom Forest, as well as the use of architectures such asFaster R-CNNandYOLO. Other project teams are also conducting experiments with differentYOLOvariants and alternative feature extractors, such asVision Transformers(ViT).

The collaborative efforts seek to achieve a flexible integration of diverse methods, which can be deployed as single-stage systems (fast and direct) or two-stage systems (slower but more precise), depending on imagery challenges and the required speed-versus-accuracy trade-off. Furthermore, Veronica emphasizes that "these innovations provide an unprecedented and practical framework to improve marine megafauna bycatch monitoring, generate more reliable species-level information, and support better-informed conservation decisions".

The initiative aims to offer users a unified, flexible, and scalable system capable of managing complex imagery within a cohesive workflow. Jordi Fornés, a research technician who contributed to the experimental analysis, underlines that "these models can be refined as more images become available and adapted to other datasets, species, and monitoring contexts".

For these studies, onboard image acquisition was carried out by vessel crews, while data collection and annotation support were provided by the company DataFish, the AI4OCEANS team and project collaborators.

Article references:
J. Fornes-Mengual, V. Nieves, “Advancing AI-based species identification for marine bycatch monitoring: Insights from faster R-CNN experiments”, Ecological Informatics (2026), https://doi.org/10.1016/j.ecoinf.2026.103960

J. Fornes-Mengual, V. Nieves, “AI-Based Image Analysis for Marine Species Monitoring: A CNN-RF Approach”,IEEE Asia-Pacific Conference on Geoscience, Electronics and Remote Sensing Technology (AGERS), Purwokerto, Indonesia, 2025, pp. 732-737, doi:10.1109/AGERS67633.2025.11446450