AI based maturity camera for farmed fish
.
Mohn Technology has developed an AI based machine vision prototype camera for Firda Seafood Group that analyses and counts the percentage of mature fish in the population.
Firda Seafood Group contacted Mohn Technology with a wish to automatically analyse the maturity of their farmed fish when it is processed at one of their ships. All the fish in the net pen is transported on a chute, and Firda Seafood Group wished to use machine vision to analyse their population here. They saw an opportunity to follow the percentage of mature fish, and thereby get a choice to act on results.
The project was started in mid January, where the fish were already starting to mature, so quick data gathering was needed. Our FRS Camera was selected as an initial platform, mounted on a frame for the chute and sent to Firda for data collection. We quickly found out that splashing water on the window would be an issue, and since the FRS camera is developed for underwater applications it had no wiper. This experience contributed to the hardware requirements for the Maturity Camera. The fast deployment of the FRS camera for data collection gave us the opportunity to work parallel on hardware, software and AI algorithm development.
The requirement for the system was that it should be delivered well in time before the next maturity cycle in the fall the year after. The system would need to fit the given chute, work under all weather and light conditions, and give the operator an accurate number of mature vs immature fish. The prototype conists of a custom camera unit with integrated high CRI lights, wiper and a ultra HD machine vision camera. It is connected to a control box and a server that is located inside the ship.
The prototype was delivered within the time frame, and during the winter we gathered additional data with higher quality (better lighting and no droplets at the window). The system is now operational, where Firda Seafood Groups experts can log on to the camera from their desk and follow the given operation in detail. The clip below shows the prototype web GUI where the user can follow the process and mautre fish count is displayed.
