A novel system for automated continuous on-farm assessment of digital dermatitis using Artificial Intelligence

Shahbaz, Ajmal; Zhang, Wenhao; Smith, Melvyn

Abstract

Digital dermatitis (DD) is a major cause of lameness in cattle, significantly affecting animal well-being and reducing productivity across the dairy industry. This paper introduces a novel system combining innovative hardware coupled with a two-stage intelligent image analysis pipeline for the early detection of DD lesions, with the goal of preventing lameness in dairy cows. By synergistically integrating image capture with foot cleaning, high-quality video of hooves are captured as cows step away from a cleaning and disinfecting footbath as part of their daily routine. Image analysis comprises two stages. An initial analysis stage for the detection of usable hoof images, accomplished through a YOLO-based ‘hoof detector’ together with filtering to remove motion blur and occlusions caused by water and mud, and a second stage ‘lesion detector and classifier’. The resulting hoof lesion detector system, the first of its kind, distinguishes between five different DD lesion stages on the cropped sole of the hooves: M1, M2, M3, M4, and M4.1, aligned with the gold-standard M- (Mortellaro) stage scoring system. Validated on real-world data automatically collected from four different working farms, the hoof detector has achieved F-scores exceeding 90%, while the lesion detector maintains F-scores over 82% for active lesions (M1 and M2). By addressing the critical issue of lameness within the dairy industry, this combined cleaning and inspection system not only demonstrates pioneering precision in hoof identification and lesion classification but also offers a promising solution for early intervention and prevention of hoof diseases in dairy cattle. To the best of our knowledge, this work represents the first instance of utilising hoof images automatically collected from real farms to apply machine learning analysis for the automated high-throughput assessment of DD lesions across five classes. A subset of the image dataset is available upon request (by contacting the authors) for research and educational purposes.

• Journal Article Type: Article
• Acceptance Date: Jul 7, 2025
• Deposit Date: Jul 10, 2025
• Journal: Smart Agricultural Technology
• Electronic ISSN: 2772-3755
• Publisher: Elsevier
• Peer Reviewed: Peer Reviewed
• Public URL: https://uwe-repository.worktribe.com/output/14687070