Underwater fish mass estimation using pattern matching based on binocular system

Abstract

Fish mass is the main information for judging growth status, regulating water quality environment, and precision feeding and grading in the process of intelligent aquaculture management activities. However, the occlusion, bending, and poor imaging angle of fish body image are still serious challenges for underwater fully automatic mass measurement. The aim of this study was to develop underwater non-contact method to automatically estimate the free-swimming fish mass based on binocular stereo vision technology. The fish body images were automatically selected and obtained by using pattern recognition method based on LabVIEW development platform during the experimental period. All the fish samples were divided into three groups according to mass (200–500 g, 500–800 g, and 800–1200 g), and then subdivided into three groups by imaging angle (orthogonal angles, greater than 45°, and less than 45°). The experiment indicated that the fish mass could be estimated using fish body area with a high coefficient of determination (R2) based on linear model. The mean relative errors between estimated and measured value were 3.37% (orthogonal angles), 4.95% (greater than 45° angles), and 16.59% (less than 45° angles). Significant difference was found in less than 45° group with p < 0.01. These findings showed that the approach put forward in this research could realize fully automatic mass estimation for underwater free-swimming fish and effectively improve the estimation robustness and efficiency.