The cost of feed is by far the highest cost factor in salmon farming. An important element during the feeding is to monitor the appetite of the fish as it may vary a lot from day to day. Traditionally this has been done by observing the behaviour of the fish at the water surface, but nowadays, an underwater camera is used to provide a better view of the fish behaviour in the feeding zone.
The project has developed a novel convolutional neural network (CNN) architecture, TridentNet, to simultaneously estimate behaviour, fish and pellet density and observation of non-fish classes based on videos from Scale Aquaculture's feeding cameras. This is a necessary step to go from purely qualitative video input to quantifying appetite related parameters.
