a pioneering project in the Pyrenees

Having conquered very different fields such as medicine, commerce, aviation, robotics or sports, artificial intelligence continues to expand exponentially in human activities. In New Aquitaine, an innovative national-scale project aims to create artificial intelligence applied to fish farming, especially trout, to improve selection, yield but also growth conditions and even mortality…

Having conquered very different fields such as medicine, commerce, aviation, robotics or sports, artificial intelligence continues to expand exponentially in human activities. In New Aquitaine, an innovative national-scale project aims to create artificial intelligence applied to fish farming, especially trout, to improve selection, yield but also growth conditions and even fish mortality.

The project Sepia (Selection by artificial intelligence for aquaculture species) was born within the walls of Viviers de Sarrance, in Pyrénées-Atlantiques: a fish farm that specializes in the production of 150 million trout eggs every year, which it sends to the four corners. of the planet. For the occasion, it joined forces with competing companies that have become partners in this development project.

The whole thing brings together a consortium of five private companies – including Landes Aqualande – and “at least three research units from Inrae and Ifremer” (the National Research Institute for Agriculture, Food and the Environment and the French Research Institute for Exploitation of the sea). Budget maintained: 3.3 million euros.

“Sepia is a collective project that we started. Among the partners, there are two trout farmers, one of marine fish and another of oysters. It is a Toulouse design office responsible for the development of artificial intelligence. File management was entrusted to Sysaaf [Syndicat des sélectionneurs avicoles et aquacoles français, NDLR], which I chair. Sepia consists of trying to apply the knowledge gained in artificial intelligence to our fish selection in the broadest sense,” explains Frédéric Cachelou, director of Viviers de Sarrance.

For a few grams more

Fish selection is at the heart of Viviers de Sarrance's business. “We have subjective selection criteria to select the most beautiful fish and other more objective criteria to measure performance. More and more, fish end up in processing plants and our industrial customers demand the highest possible performance. At the moment, we can get around 500 grams of fillets from one kilo of trout. So increasing to even 550 grams would already be a game changer. »

Until now, this selection was done manually. “For each selection generation we take 1,000 fish, weigh them and then make the fillets which we weigh. It is a painstaking process that takes time. Artificial intelligence should greatly facilitate this task,” explains Viviers de Sarrance director.

Better than the human eye

To help the AI ​​understand what types of fish to choose, the Sepia project involves a phase of “feeding” the machine, bombarding it with data. “We'll take thousands of photos of trout in all directions and tell the machine what we want to keep and what we don't want to keep. She will learn on her own as we give her our criteria. In the end, we hope to save time, accuracy and rigor. »

In particular, it is expected that AI analysis will be much more effective in terms of the aesthetic selection of fish. “From one human eye to another, judgment can vary significantly. Artificial intelligence will not have this problem,” says Frédéric Cachelou.

Hardier fish

The machine could be useful in other applications, explains Émilien Segret, head of the reproduction program at Viviers de Sarrance. “We also intend to develop a mathematical formula to capture, in the fish genome, what constitutes disease resistance. This could make it possible to reduce mortality and we could thus reduce the supply of antibiotics. »

The last aspect sought through the Sepia project: achieving facial recognition of fish to know exactly what food they have ingested. “The goal is to have fish that better convert the food we give them, so they have less release of nitrogen and phosphate materials,” the researcher continues.

Winner of the France 2030 investment plan, Sepia should receive a total aid of 2.6 million euros out of the 3.3 million (a percentage expected to increase in particular due to inflation) required for its development. “We will start the project as soon as possible in 2024. It should be completed within 2027 and we hope to apply it to genetic selection from the end of 2027,” announces Frédéric Cachelou.

Leave a Reply

Your email address will not be published. Required fields are marked *