Vision detection for early signs of DD lesions and lameness within dairy cattle

Shahbaz, Ajmal; Zhang, Wenhao; Smith, Melvyn

Abstract

Digital dermatitis stands as a primary cause of lameness in dairy cows, significantly impacting various facets of productivity. This paper proposes a two-stage vision system aimed at early detection of digital dermatitis (DD) lesions, ultimately preventing lameness in dairy cows. The first stage involves the detection of dairy cow hooves, achieved through a YOLO-based hoof detector. The detected hooves undergo a meticulous filtering process to gather high-quality data exclusively. These selected hooves are then cropped from the original images to train the lesion detector. The lesion detector distinguishes between six different classes on the cropped hooves: lesion, M1, M2, M3, M4, and M4.1. Initial results demonstrate the hoof detector achieving precision rates above 90%, while the lesion detector maintains precision rates surpassing 70%. In addressing the critical issue of lameness, this two-stage vision system not only showcases impressive precision in hoof and lesion detection but also highlights the potential for early intervention and prevention in the realm of dairy cow health. The robust performance of the detectors lays the groundwork for further advancements in proactive veterinary care within the dairy industry.

• Presentation Conference Type: Conference Paper (published)
• Conference Name: IEEE 22nd World Symposium on Applied Machine Intelligence and Informatics
• Start Date: Jan 25, 2024
• End Date: Jan 27, 2024
• Acceptance Date: Dec 5, 2023
• Deposit Date: Jan 30, 2024
• Keywords: Index Terms-Computer vision; machine learning; artificial intelligence; object detection; Lameness; Digital Dermititis
• Public URL: https://uwe-repository.worktribe.com/output/11641571