Fish production by the aquaculture industry has steadily increased to provide alternative fish resources. To overcome difficulties associated with a manual labor-intensive farming technologies and create a more systematic aquaculture management system that is compatible with the oncoming fourth industrial evolution, it is important to automate some aspects of the aquaculture industry such as water quality detection, automatic feeding systems, and real-time underwater monitoring. Despite of technologies have already been adapted to monitor physical conditions, those for automatic monitoring of fish behavior are still needed for real-time fish condition assessment. In this study, we applied the YOLO deep learning algorithm to detect abnormal swimming behavior of rock bream Oplegnathus fasciatus, based on fish movement data. We recorded images of rock bream before and after adding an anesthetic (MS-222) or the replacement of seawater with fresh water and then evaluated the ability of algorithm to detect fish displaying abnormal behavior. The proposed algorithm showed a high accuracy (88.1%) in discriminating normal and abnormal rock bream behavior. We conclude that artificial intelligence-based detection of abnormal behavior can be applied to develop an automatic biomanagement system for use in the aquaculture industry.
The collected data included 10,110 rock bream images obtained from the video recording. Among these, 210 images were randomly selected and divided these into a training dataset containing 168 images and a test dataset containing 42 images. Images of rock bream swimming upright and lying on their side were considered to exhibit normal and abnormal swimming, respectively