A team from the Institute of Automation and Industrial Informatics at the Universitat Politècnica de València has managed to replicate at sea—in the Mediterranean, Adriatic, and eastern Atlantic—an innovative automatic tuna counting system. The result is part of the REM-BFT project (Remote Electronic Monitoring on bluefin tuna processing vessels), funded by the European Union, which concluded last summer, addressing a long-standing demand from fishing companies and the International Commission for the Conservation of Atlantic Tunas. The system enables better control of fishing and the size of captured specimens.
Over the past three years, the ai2 research team has worked to improving its tuna counting system, a project they have been developing for 15 years. The goal: to replicate it at sea. This system measures not only the number of specimens but also their biomass. During tuna fishing, two fish transfers occur: the first at sea, from the purse seine cage to the transport cage; the second, closer to the coast, from the transport cage to the fattening farms.
“Traditionally, specimen control is carried out by reviewing recordings and manually marking the tuna by an operator during the second transfer. Being able to replicate this measurement system at sea allows companies to calculate the fishing quota during the capture process and not weeks later when they arrive at the farm,” explains Pau Muñoz, researcher at the ai2 Institute.
“Being able to replicate this measurement system at sea allows companies to calculate the fishing quota during the capture process and not weeks later when they arrive at the farm,” explains Pau Muñoz, researcher at the ai2 Institute.
The stereoscopic computer vision system based on deep learning techniques developed by ai2 provides reliable biometric values for captured tuna while swimming, improving weight estimation at capture far more accurately than traditional methods.
The technology combines diverse knowledge and techniques: cameras, optics, sensors, underwater housings, high-performance GPUs for processing, segmentation techniques, classification, machine learning, and statistics.
“For 15 years, we have advanced the techniques used to obtain information about bluefin tuna individuals from images collected while they swim freely in fattening pools, facing challenges inherent to developing automatic systems that operate in natural (uncontrolled) environments with free-moving animals—such as multiple individuals appearing in the image, distinguishing between a snout and a tail, etc.,” notes Gabriela Andreu, principal researcher of the project at the ai2 Institute of UPV.
Studies in Spain, Croatia, and Portugal
The major step forward has been transferring these studies from fattening farms to open sea. To achieve this, several system tests were conducted over the past three years. The first took place in the western Mediterranean with the collaboration of Balfegó. The next was in Croatia, in the Adriatic. The latest tests were carried out this past summer in the Almadrabas of the eastern Atlantic, in Portugal.
“The scrutiny differs in each location, especially in the Adriatic, as the tuna are of different sizes and enter more densely grouped,” explains Muñoz. “We have verified that the measurements provided by the system are highly accurate. We continue working to refine specimen counting as much as possible, but the project now concludes successfully,” the researcher assures.
Other species
In parallel, this ai2 team continues working on other biomass estimation projects for marine species, in collaboration with the Torre la Sal Aquaculture Institute (IATS-CSIC) and the company Avramar, focusing on species such as gilthead seabream and European seabass.