Developing and Assessing the Performance of a Machine Learning Model for Analyzing Drinking Behaviors in Minipigs for...

Area of Science:

• Animal Behavior
• Machine Learning
• Computer Vision

Background:

• Monitoring experimental animals is crucial for ethical, scientific, and financial reasons.
• Conventional observation methods for animal behavior are subjective and time-consuming.
• Automated monitoring systems using machine learning offer a promising alternative.

Purpose of the Study:

• To validate and assess a machine learning model for analyzing drinking behavior in minipigs.
• To develop and test a novel, vision-based monitoring system for on-site video analysis.
• To evaluate the performance of a computer vision system for detecting drinking events.

Main Methods:

• A dataset of 5297 images was used to train a YOLOv11n object detection model.
• The model identified pig heads and water faucets to detect drinking events.
• A low-cost Raspberry Pi system enabled real-time video analysis and on-site processing.

Main Results:

• The object detection model achieved over 97% accuracy.
• Manual validation confirmed high overall accuracy (99.7%), precision (99.7%), recall (99.2%), and F1-score (99.5%).
• Analysis of drinking patterns revealed bimodal behavior and inter-pig variability.

Conclusions:

• A low-cost, computer vision system can effectively monitor drinking behavior in individually housed experimental pigs.
• This technology supports earlier detection of potential illness in animals.
• The system demonstrates feasibility for automated, objective animal welfare monitoring.