Deep learning enhanced thermographic modeling for early and precise mastitis detection in Sahiwal cows

Abstract

Mastitis, a multifactorial production disease, poses a significant challenge to dairy farming, necessitating the adoption of integrated and precision-based diagnostic approaches. This study explores the potential of thermal imaging combined with deep learning to enhance mastitis detection in lactating dairy cows. In this study, thermal images of the udder region of Sahiwal cows were captured using a handheld thermal camera and analyzed to classify udder quarters as healthy, Sub-clinical Mastitis (SCM), and Clinical Mastitis (CM). The classification was based on California Mastitis Test (CMT) scores, Somatic Cell Count (SCC) values, and thermal image analysis. Further, Convolutional Neural Network (CNN) models were developed to distinguish between healthy udder quarters and those affected by CM or SCM. The CNN model differentiating healthy quarters from CM achieved training, validation, and testing accuracies of 99 %, with precision, recall, and F1-score all at 0.99. Similarly, the model distinguishing healthy quarters from SCM recorded training and validation accuracies of 89 % and 85 %, respectively, while testing results showed an accuracy of 84 %, a precision of 0.87, a recall of 0.79, and an F1-score of 0.83. These findings highlight the potential of CNN-based thermal imaging for accurate mastitis detection, contributing to advancements in precision dairy farming and livestock health management.