Real-time digital dermatitis detection in dairy cows on Android and iOS apps using computer vision techniques

Agam Dwivedi ¹, Marlee Henige ¹, Kelly Anklam ¹

⁰School of Veterinary Medicine, University of Wisconsin-Madison, Madison, WI 53706, USA.

Translational Animal Science +

|

February 17, 2025

View abstract on PubMed

Summary

This study developed mobile apps using computer vision for early detection of digital dermatitis (DD) in cows. The iOS app showed moderate agreement with human investigators, aiding timely treatment and improving animal welfare.

Area Of Science

Veterinary Medicine
Computer Vision
Machine Learning

Background

Digital dermatitis (DD) is a significant cause of lameness in dairy cattle, impacting animal welfare and farm economics.
Early detection of DD lesions is crucial for effective treatment and preventing herd-wide outbreaks.
Mobile applications offer a potential solution for accessible, real-time DD detection on farms.

Purpose Of The Study

To deploy computer vision models for real-time detection of digital dermatitis (DD) lesions in cows using Android and iOS mobile applications.
To evaluate the performance of these models in identifying different stages of DD lesions.
To facilitate early intervention and management strategies for DD on dairy and beef farms.

Main Methods

Transfer learning was applied to a YOLOv5 model architecture using pre-trained COCO-128 weights for DD image data.
The model was trained on 363 images across 5 lesion classes (M0, M4H, M2, M2P, M4P) and augmented for robustness.
Models were converted to TFLite (Android) and CoreML (iOS) formats, with quantization for optimized inference speed.

Main Results

The DD detection model achieved a mean average precision (mAP) of 0.95 on the test dataset.
iOS devices showed moderate agreement (Cohen's kappa = 0.57) with human investigators across 5 lesion classes.
A 2-class model (lesion vs. non-lesion) achieved higher agreement, with iOS reaching kappa = 0.74 and Android kappa = 0.65. iOS also demonstrated faster inference times (20 ms vs. 57 ms).

Conclusions

Mobile applications powered by computer vision can effectively support the early detection of digital dermatitis in cattle.
The developed iOS application demonstrated promising performance, surpassing existing apps in detection accuracy and speed.
This technology can empower farmers with tools for timely DD management, improving herd health and welfare.

Keywords:

YOLO animal welfare app deployment cattle computer vision deep learning