Video-Based Automated Lameness Detection for Dairy Cows.

Area of Science:

• Veterinary Medicine
• Data Science
• Animal Science

Background:

• Lameness in cattle incurs significant treatment costs, ranking second among common diseases.
• Current visual lameness detection is time-consuming, labor-intensive, and prone to subjective discrepancies.
• There is a need for objective, automated methods for lameness assessment in dairy herds.

Purpose of the Study:

• To develop fully automated, end-to-end data science methods for video-based lameness assessment in dairy cows.
• To introduce and evaluate a custom 7-point locomotion scoring system for lameness severity.
• To compare machine learning and deep learning approaches for lameness detection.

Main Methods:

• Utilized video recordings of 832 cows with varying lameness degrees.
• Employed deep learning for cow detection and anatomical point identification in video frames.
• Analyzed spine curvature, head position, and leg distances using expert systems, machine learning, and deep learning.

Main Results:

• Developed a 7-point locomotion scoring system inspired by the Sprecher scale.
• Achieved high accuracy in detecting binary lameness (healthy vs. lame), reaching 0.821 and 0.872 with multiple locomotion features.
• Demonstrated that automated analysis approaches expert-level performance.

Conclusions:

• Fully automated, video-based lameness detection using data science is feasible and effective.
• The developed methods provide an objective and efficient alternative to traditional lameness assessment.
• This technology has the potential to reduce economic losses associated with cattle lameness.