A Deep Learning Lightweight Model for Real-time Captive Macaque Face Recognition Based on an Improved YOLOX Model

Automated behavior monitoring of macaques has the potential to advance biomedical research and animal welfare. However, reliably identifying individual macaques in group settings remains a significant challenge. This study introduces ACE-YOLOX, a lightweight facial recognition model tailored for captive macaques. ACE-YOLOX integrates Efficient Channel Attention (ECA), Complete Intersection over Union loss (CIoU), and Adaptive Spatial Feature Fusion (ASFF) into the YOLOX framework to enhance prediction accuracy and reduce computational complexity. These approaches enable effective multiscale feature capture. Evaluated on a dataset of 179,400 labeled facial images from 1,196 macaques, ACE-YOLOX outperforms classical object detection models and demonstrates real-time processing capability. Additionally, we developed an Android application to deploy ACE-YOLOX on smartphones, enabling on-device, real-time recognition. Our experimental results highlight ACE-YOLOX’s potential as a non-invasive tool for macaque identification, offering an important foundation for future studies in macaque facial expression recognition, cognitive psychology, and social behavior.