Automating hock wound detection in dairy cattle.

Area of Science:

• Animal Welfare Science
• Veterinary Medicine
• Artificial Intelligence in Agriculture

Background:

• Hock scoring is a critical welfare indicator in dairy cattle, assessing hock condition for injuries and lesions.
• Inconsistent manual scoring by observers can compromise the reliability of this vital welfare assessment.
• Existing methods highlight a need for more objective and consistent approaches to evaluate dairy cow hock health.

Purpose of the Study:

• To quantify the inconsistency of manual hock scoring in dairy cattle.
• To explore the potential of artificial intelligence for automating hock score detection.
• To develop a more reliable and objective method for assessing dairy cattle hock health.

Main Methods:

• Two studies were conducted to measure manual scoring reproducibility and repeatability.
• Repeatability was assessed using a weighted Cohen's kappa metric.
• A U-net semantic segmentation algorithm was employed for automated wound detection on cow hocks.

Main Results:

• Manual hock scoring demonstrated significant inconsistency.
• Manual scoring was found to be more consistent than video scoring.
• The U-net algorithm successfully detected hock wounds, indicating potential for automation.

Conclusions:

• Variability in manual hock scoring necessitates more objective assessment methods.
• Artificial intelligence-driven automation can reduce observer bias and improve scoring consistency.
• Automated hock scoring promises enhanced dairy cattle welfare through efficient and accurate health monitoring.