Intelligent Recognition of Chinese Pangolin Stereotyped Pacing

Pangolins, such as the critically endangered Chinese and Malayan species, often develop stereotypic behaviors in captivity. To automate the detection of these behaviors, we propose a video recognition method using an improved YOLO v8n-CBAM model. It detects pangolins and maps their image coordinates to real-world locations, generating movement trajectories. These are segmented and analyzed by a ResNet18 model to identify stereotypic behavior. Our improved YOLO model achieves 99.93% recall and 95.63% mAP@0.50:0.95. Trajectory distance accuracy reaches 98.48%, and stereotypic behavior periods are detected with an average accuracy of 96.37%. This automated approach provides a scientific basis for improving pangolin husbandry and conservation management.