Automated Cow Body Condition Scoring Using Multiple 3D Cameras and Convolutional Neural Networks

Area of Science:

• Animal Science
• Computer Vision
• Machine Learning

Background:

• Body condition scoring (BCS) is crucial for dairy cow health assessment.
• Automated BCS enhances efficiency and consistency over manual methods.
• Convolutional Neural Network (CNN) models with depth cameras are common for automated BCS.

Purpose of the Study:

• To evaluate the performance impact of ensemble modeling using data from three depth cameras for automated BCS.
• To determine optimal camera configurations balancing computational cost and accuracy.
• To assess real-world performance on embedded platforms.

Main Methods:

• Trained three independent CNN models using data from three depth cameras positioned around a cow.
• Employed ensemble modeling to combine predictions from individual CNN models.
• Evaluated model performance and computational cost on embedded systems.

Main Results:

• Ensemble modeling with three depth cameras significantly improved automated BCS accuracy compared to single-camera approaches.
• Identified specific camera combinations offering a favorable trade-off between computational cost and accuracy.
• Demonstrated the practical feasibility of these models on embedded platforms.

Conclusions:

• Ensemble modeling with multi-camera depth data enhances automated BCS accuracy in dairy cows.
• The study provides insights into optimizing automated BCS systems for practical dairy farm applications.
• This research contributes to more efficient and accurate dairy herd management through advanced AI.