Fish School Behaviour Classification for Optimal Feeding Using Rose Diagram

As the global demand for seafood increases, sustainable aquaculture, which does not cause overfishing or damage to the marine environment compared to traditional natural fishing, is gaining attention.In aquaculture, it has become important to efficiently raise fish with less feed by optimizing the timing and amount of feeding. Therefore, we developed a system that quantifies fish activity from captured images and calculates the optimal feeding time and feeding amount. The system uses a method called optical flow to vectorize the amount of activity of schools of fish in images and uses the numerical data from the rose diagram as an indicator. We conducted a verification of the indicator's effectiveness by quantitating the eating and not eating movements during feeding. The performance of these quantified indicators was then evaluated using a machine learning method known as support vector classification. We confirmed that the proposed indicator improved the accuracy of the classification of the two activities compared to the previous indicator.