Machine-Learning-Based Activity Tracking for Individual Pig Monitoring in Experimental Facilities for Improved Animal...

Area of Science:

• Animal Welfare Science
• Machine Learning Applications
• Agricultural Technology

Background:

• Current methods for monitoring experimental animal welfare rely on subjective and time-consuming physical observations.
• Objective and continuous monitoring is crucial for detecting early signs of animal misthriving in research settings.

Purpose of the Study:

• To develop and validate a machine learning model for objective monitoring of experimental pig posture and activity levels.
• To create a video-based pig tracking technology for enhanced animal welfare assessment.

Main Methods:

• A prototype system utilizing a microcomputer and ceiling-mounted camera was developed for live video recording.
• A combined machine learning model integrating Ultralytics YOLOv8n for object detection and Lucas-Kanade sparse optical flow for movement analysis was employed.
• The model was trained on images from experimental pigs in single-pig pens to classify postures (lying, standing, walking).

Main Results:

• The developed model successfully classified the posture and activity levels of individual pigs with high accuracy.
• Validation demonstrated a 90.66% accuracy, 90.91% precision, and 90.66% recall.
• A correlation coefficient of 84.53% was achieved when compared to observed ground truth.

Conclusions:

• Machine learning offers a viable and objective method for continuous monitoring of experimental animals.
• This technology has the potential to significantly improve animal welfare in research by enabling early detection of distress or illness.
• The validated pig tracking system provides a reliable tool for researchers focused on animal welfare.