Rats Behavior Recognition Based on Improved YOLOv8

In the domains of neuroscience and pharmacology, rodent models are crucial for deepening our understanding of neurobiological processes. Traditional behavioral recognition methods, however, encounter issues such as high costs and suboptimal accuracy. This research presents a pioneering method that employs Deformable Attention to enhance rat behavior recognition on an observation platform. By integrating Deformable Convolutional Networks (DCNs) into the YOLOv8 algorithm, the approach amplifies detection precision and robustness. We performed experiments utilizing six hours of video data from rats navigating an eight-arm maze, categorizing behaviors including walking, sniffing, barbering, eating and so on. The refined model showed a substantial increase in mean Average Precision (mAP), from 0.776 to 0.990, along with a smoother and more stable loss curve, indicating enhanced predictive performance. This study demonstrates the feasibility of leveraging standard cameras and sophisticated machine learning techniques for effective and precise behavior analysis under limited resource conditions.