SATSN: A Spatial-Adaptive Two-Stream Network for Automatic Detection of Giraffe Daily Behaviors.

Area of Science:

• Animal Behavior
• Computer Vision
• Artificial Intelligence

Background:

• Giraffe daily behaviors are key indicators of health and well-being, especially in captive settings.
• Automated detection systems are crucial for scientific management and welfare improvement in zoos.
• Repetitive behaviors in captive giraffes can signal underlying welfare concerns.

Purpose of the Study:

• To develop an efficient and accurate automated multi-behavior detection system for giraffes.
• To improve the scientific management and welfare assessment of giraffes in zoos.
• To leverage advanced deep learning models for behavioral analysis.

Main Methods:

• Utilized YOLO11-Pose for giraffe detection and keypoint estimation (mouth).
• Employed Observation-Centric SORT (OC-SORT) for individual giraffe tracking.
• Developed a spatial-adaptive two-stream network (SATSN) with a video transformer and Temporal Attention (TA) module for behavior classification.

Main Results:

• Achieved a mean average precision (mAP) of 93.99% for detecting licking, walking, standing, and eating behaviors.
• Demonstrated strong detection performance and generalization capability on a custom giraffe behavior dataset.
• Validated the effectiveness of the proposed YOLO11-Pose and SATSN integrated approach.

Conclusions:

• The proposed automated system provides robust support for multi-behavior detection and well-being assessment in giraffes.
• This technology lays a foundation for intelligent behavioral monitoring systems in zoological settings.
• The method enhances the ability to monitor and improve giraffe welfare through accurate behavioral analysis.