Deep learning and image processing for high-throughput udder phenotyping in dairy cows.

Area of Science:

• Animal Science
• Computer Vision
• Agricultural Engineering

Background:

• Udder conformation is vital for dairy cow health and productivity, especially in Automated Milking Systems (AMS).
• Accurate phenotyping is essential for genetic selection, health monitoring, and welfare assessments.
• Existing methods for udder phenotyping can be labor-intensive and subjective.

Purpose of the Study:

• To develop an automated pipeline for extracting udder phenotypes from depth images.
• To evaluate the association between udder conformation traits and milk production performance.
• To enable high-throughput phenotyping for improved dairy cattle management.

Main Methods:

• Depth videos of 150 Holstein cows were recorded in an AMS environment.
• Convolutional Neural Networks (CNNs) were trained for image segmentation and keypoint detection.
• The pipeline integrated CNNs with image processing for udder segmentation, teat localization, and phenotype extraction.

Main Results:

• The automated pipeline achieved high accuracy in udder segmentation (IoU = 0.93) and teat localization (NED = 0.061).
• Extracted phenotypes included udder/quarter circularity, eccentricity, surface area, volume, and teat dimensions.
• Larger udder area was significantly associated with higher milk yields (r = 0.55).
• Increased parity correlated with larger udder perimeter, longer teats, and greater teat distance.

Conclusions:

• The developed pipeline offers a robust and efficient solution for high-throughput udder phenotyping.
• Udder conformation traits are valuable indicators of milk production and can inform breeding strategies.
• This technology has significant implications for dairy cattle breeding, health management, and animal welfare.