Deep-learning classification of teat-end conditions in Holstein cattle

Area of Science:

• Veterinary Medicine
• Artificial Intelligence
• Animal Science

Background:

• Mastitis is a significant concern in dairy farming, impacting animal health and farm economics.
• Current methods for assessing teat-end conditions, crucial for mastitis prevention, can be subjective and labor-intensive.
• Optimizing deep learning for classifying teat-end hyperkeratosis is essential for early detection and intervention.

Purpose of the Study:

• To evaluate the efficacy of deep learning models in classifying teat-end conditions in dairy cows.
• To assess the potential of transfer learning approaches using ImageNet models for this classification task.
• To determine the accuracy and reliability of image-based assessment for hyperkeratosis detection.

Main Methods:

• A dataset of 1426 digital images of dairy cow udders was utilized.
• Teat-end hyperkeratosis was assessed using a four-point scale.
• Transfer learning with eight different ImageNet models was employed for image classification.
• Data was partitioned into training (70%), validation (15%), and testing (15%) sets.

Main Results:

• Eight ImageNet models achieved high classification accuracy, ranging from 80.3% to 86.6%.
• Areas under the receiver operating characteristic curves (AUC) for different classifications (normal, smooth, rough, very rough) ranged from 0.825 to 0.999.
• The study highlights the potential of deep learning but indicates a need for more training data.

Conclusions:

• Deep learning models demonstrate significant potential for accurately classifying dairy cow teat-end conditions.
• This image-based classification method can aid farmers in reducing intramammary infections and antimicrobial use.
• Further development with larger datasets is recommended to optimize deep learning for mastitis prevention strategies.