state-of-the-art technology to minimize accidental capture

Bycatch accounts for approximately 40% of global fish catches and poses a major threat to marine biodiversity. It particularly affects megafauna, such as turtles, marine mammals, and sharks. The AI4OCEANS research team, led by researcher Verónica Nieves at the University of Valencia, has developed hybrid AI frameworks to automatically identify species caught by mistake, using images taken directly on board fishing vessels.

The researchers combine different AI architectures (such as convolutional neural networks—CNN, Faster R-CNN, YOLO, and Random Forest) to create flexible systems. The system allows users to choose between single-stage models (faster, for real-time processing) or two-stage models (slower but more accurate) to adapt to image quality and monitoring requirements. These models are designed to transform complex images into reliable biodiversity information, even when photo quality is poor. This work is part of the European REDUCE project (HORIZON program), which aims to improve bycatch estimates and support conservation decisions based on accurate data.

This research proposes a scalable and practical method that uses AI to more effectively monitor bycatch of marine wildlife, thereby enabling better protection of vulnerable species through more reliable monitoring.