Identification methodology of special behaviors for fish school based on spatial behavior characteristics

• Published in: Computers and electronics in agriculture Vol. 185; p. 106169
• Main Authors: Yu, Xiaoning, Wang, Yaqian, An, Dong, Wei, Yaoguang
• Format: Journal Article
• Language: English
• Published: Amsterdam Elsevier B.V 01.06.2021; Elsevier BV
• Subjects: Aquaculture; Behavior; Corner detection; Feature extraction; Fish; Fish welfare; Image segmentation; Kinetic energy; Model accuracy; Optical flow (image analysis); Simulation feature point selection; Spatial behavior characteristics; Special behaviors detection; Simulation feature point selection; Spatial behavior characteristics; Fish welfare; Special behaviors detection; Aquaculture
• ISSN: 0168-1699; 1872-7107
• DOI: 10.1016/j.compag.2021.106169

•A novel video feature point extraction method was applied to extract the special behavior features of fish school.•An integrated solution based on local outlier factors was constructed to identify special behaviors.•Simulation feature point selection scheme can magnificently obtain effective results in monitoring and detection process. The identification of special behaviors for fish school is an effective approach to monitor fish welfare and improve the informatization level of aquaculture. In order to monitor and detect special behaviors of fish school, such as stress and feeding behavior, a scheme based on simulation feature point selection (SFPS) is proposed. In this scheme, feature points extraction and special behaviors identification are integrated seamlessly. In order to avoid the issues of poor stability and complex operation led by foreground segmentation, an image is divided into several sub-images with the help of sliding window. Feature points and velocities of sub-images are then obtained by the Harris corner detection and Lucas-Kanade optical flow. Moreover, a model coupled normal behavior characteristic matrix and the detection of special behaviors is designed to identify special behaviors of fish school. Furthermore, comparisons among SFPS, Single Gaussian Model (SGM) and the kinetic energy model are carried out. The results show that the scheme can effectively monitor the special behaviors of fish school, and the detection accuracy of feeding behavior can achieve to 96.02%. Comparing with SGM and the kinetic energy model, the accuracy of stress behavior and feeding behavior for SFPS scheme are improved by 13.25%, 4.21%, 30.24%, and 4.32% respectively. Also, the operation time is reduced by 98.73% without the foreground segmentation. The SFPS scheme proposed in this paper is not only conducive to the development of special behaviors identification technology for fish school, but also provides a feasible solution for the application of this technology in practical engineering.