Individual Identification of Medaka, a Small Freshwater Fish, from the Dorsal Side Using Artificial Intelligence

Abstract

Individual identification is an important ability for humans and perhaps also for non-human animals to lead social lives. It is also desirable for laboratory experiments to keep records of each animal while rearing them in mass. However, the specific body parts or the acceptable visual angles that enable individual identification are mostly unknown for non-human animals. In this study, we investigated whether artificial intelligence (AI) could distinguish individual medaka, a model animal for biological, agrarian, ecological, and ethological studies, based on the dorsal view. Using Teachable Machine, we took photographs of adult fish (n = 4) and used the images for machine learning. To our surprise, the AI could perfectly identify the four individuals in a total of 11 independent experiments, and the identification was valid for up to 10 days. The AI could also distinguish eight individuals, although machine learning required more time and effort. These results clearly demonstrate that the dorsal appearances of this small spot-/stripe-less fish are polymorphic enough for individual identification. Whether these clues can be applied to laboratory experiments where individual identification would be beneficial is an intriguing theme for future research.

Keywords

• medaka
• individual identification
• artificial intelligence
• Teachable Machine