Automated dairy cattle body condition score using side-view images and deep learning

Area of Science:

• Animal Science
• Computer Science
• Artificial Intelligence

Background:

• Traditional body condition score (BCS) assessment in dairy cows is labor-intensive and subjective.
• Inadequate BCS monitoring, especially in smallholder farms, can lead to overconditioning and metabolic issues like ketosis.
• Automated BCS systems are needed for efficient, objective, and large-scale dairy herd health management.

Purpose of the Study:

• To develop and validate a fully automated, regression-based system for estimating dairy cow BCS from single side-view images.
• To create a noninvasive, cost-effective, and infrastructure-light tool for precision dairy farming.
• To ensure the AI model learns biologically relevant features for accurate BCS prediction.

Main Methods:

• A two-stage deep learning approach was used: YOLOv11n for object detection and cropping, followed by a regression model for BCS prediction.
• The system was trained and validated on 3,208 side-view images from 211 Holstein cows under real-world conditions.
• A stratified 5-fold cow-level cross-validation and explainable AI techniques were employed to ensure unbiased evaluation and model interpretability.

Main Results:

• The automated system achieved a Mean Absolute Error (MAE) of 0.41 and a Pearson Correlation Coefficient of 0.62, comparable to inter-assessor variability.
• Object detection achieved a mean Average Precision (mAP) of 99.5%, effectively isolating the bovine region of interest.
• Explainable AI confirmed the model focused on key anatomical landmarks (tailhead, hooks, pins, ribs), indicating biologically relevant feature learning.

Conclusions:

• The developed automated BCS system is a viable, noninvasive tool for precision dairy farming, enabling early detection of negative energy balance and timely nutritional interventions.
• The system's performance, validated against expert consensus and explainability analysis, demonstrates its potential for improving animal welfare and farm productivity.
• The study provides a proof-of-concept with a deployable pipeline, paving the way for robust, edge-compatible BCS assessment in dairy herds.