Individual identification of Japanese macaques (Macaca fuscata) using a face recognition system and a limited number of learning images

Abstract

Individual identification is an important technique in animal research that requires researcher training and specialized skillsets. Face recognition systems using artificial intelligence (AI) deep learning have been put into practical use to identify in humans and animals, but a large number of annotated learning images are required for system construction. In wildlife research cases, it is difficult to prepare a large amount of learning images, which may be why systems using AI have not been widely used in field research. To investigate the development of a system that identifies individuals using a small number of learning images, we constructed a system to identify individual Japanese macaques (Macaca fuscata yakui) with a low error rate from an average of 20 images per individual. The characteristics of this system were augmentation of data, simultaneous determination by four individual identification models and identification from a majority of five frames to ensure reliability. This technology has a high degree of utility for various stakeholders and it is expected that it will advance the development of individual identification systems by AI that can be widely used in field research.