The overall objective of this project is to develop advanced Computer Vision (VxC) techniques based on Deep Learning and Convolutional Neural Networks (CNN) applicable to underwater natural environments to biometrically characterize populations of bluefin tuna automatically. The specific objectives are:

• Implement Deep Learning algorithms and appropriate CNN networks to work with natural environments and species, characterized by a high degree of variability.
• Evaluate the detection and characterization of tuna in stereoscopic videos acquired in real conditions with the developments proposed in Deep Learning and compare them with those obtained with classical processing techniques. Specifically, the aim is to achieve robust mechanisms against the high intrinsic variability of natural and underwater environments (luminosity, water turbidity, density of individuals, etc.), optimizing both the number of tuna detected and the response time.
• Generalize the procedure so that the system is capable of characterizing the biometrics of tuna regardless of the position and orientation in space of the cameras and fish.
• Apply the proposed developments to the detection and tracking of individuals of bluefin tuna in transfers between cages.
• Study the feasibility of extrapolating techniques for the identification and characterization of other marine species.