Face recognition of a Lorisidae species based on computer vision

Abstract

The slow loris is a group of endangered small arboreal primates. In recent decades, illegal hunting and habitat degradation have led to a sharp decline in wild populations. The individuals are challenging to be identified both in captive and natural environments due to their cryptic nocturnal behavior. Computer vision has emerged as a new approach to face recognition for domestic and wild animals. We used a YOLOv5 +U-Net+VGG framework to realize the face identification of Bengal slow lorises (Nycticebus bengalensis) based on 1480 images of 30 individuals housed in Dehong Wildlife Rescue Center, China. This is the first human-annotated face image dataset of Lorisidae primates. The accuracy rate of this deep learning model set for face recognition reached 96.27 %. The results indicated that computer vision and deep learning technology could be used in individual identification of slow lorises, contributing to further study and conservation of this endangered taxa.