Intelligent Recognition of Chinese Pangolin Stereotyped Pacing

Area of Science:

• Animal Behavior
• Computer Vision
• Conservation Science

Background:

• Critically endangered pangolin species, including Chinese and Malayan pangolins, exhibit stereotypic behaviors in captivity.
• Stereotypic behaviors can indicate poor welfare and stress in captive animals, necessitating monitoring.
• Current methods for detecting these behaviors are often manual and labor-intensive.

Purpose of the Study:

• To develop and validate an automated video recognition system for detecting stereotypic behaviors in pangolins.
• To improve the accuracy and efficiency of monitoring pangolin welfare in captive environments.
• To provide a data-driven foundation for enhancing pangolin husbandry and conservation strategies.

Main Methods:

• An improved You Only Look Once version 8 nano-CBAM (YOLO v8n-CBAM) model was utilized for pangolin detection and coordinate mapping.
• Pangolin movement trajectories were generated by mapping image coordinates to real-world locations.
• A Residual Network 18 (ResNet18) model was employed to segment and analyze movement trajectories for stereotypic behavior identification.

Main Results:

• The improved YOLO v8n-CBAM model achieved high performance with 99.93% recall and 95.63% mAP@0.50:0.95.
• Trajectory distance accuracy reached 98.48%.
• Stereotypic behavior detection accuracy averaged 96.37%.

Conclusions:

• The automated video recognition system effectively detects stereotypic behaviors in pangolins.
• This technology offers a reliable and objective method for assessing pangolin welfare in captivity.
• The findings support the improvement of pangolin husbandry and conservation management through automated monitoring.