Automatic underwater weight estimation of omega perch by computer vision

The aim of this study was to test and evaluate a computer vision technique that estimates the weight of Omega Perch swimming freely in a tank of a recirculation aquaculture system. A set of 105 images of fish were captured out of the water in order to build up a relationship between the fish shape and its weight. Regression analysis revealed that the coefficient of determination (R2) between area and weight is 0.98. Based on this model the mean predicted error weight is 9±7% when compared with the value measured by a weighing scale. This model was then used to estimate the biomass of 15 fish that were moving freely through an underwater corridor where a camera was installed. The results revealed a 10±5% estimation error when compared with the value measured by a weighing scale. Therefore the proposed method might be useful for continuously and automatically estimating the biomass of freely swimming fish without causing stress or damage to them.status: publishe