Advanced mastitis detection in Bos indicus cows: A deep learning approach integrated with infrared thermography

Mastitis is a significant challenge in the global dairy industry, impacting animal welfare, milk quality, and production efficiency. This study proposes infrared thermography (IRT) combined with deep learning algorithms for mastitis detection in Tharparkar cattle (Bos indicus), a breed known for its resilience to harsh climates. A total of 7223 udder thermograms were collected from quarters classified as healthy, subclinical mastitis (SCM), and clinical mastitis (CM) across seasons. Thermographic imaging, the California Mastitis Test (CMT), and somatic cell count (SCC) were used to assess udder health. Thermographic variables were evaluated using receiver operating characteristic (ROC) analysis to establish threshold values. Convolutional neural networks (CNNs'), alongside ResNet-50 and VGG16 architectures, were applied to classify udder health statuses. The results revealed significant increases (p < 0.01) in the udder and teat skin surface temperatures (USST and TSST) in SCM and CM quarters compared to healthy ones, with the most substantial changes in winter. USST and TSST correlated strongly with CMT scores and log₁₀SCC values (p < 0.01). CNN models outperformed others, achieving training and validation accuracies of 0.98 and 0.93 for Normal vs. Clinical, and 0.98 and 0.90 for Normal vs. Sub-clinical cases. This study demonstrates the potential of integrating IRT with AI-driven diagnostic models to provide non-invasive, accurate mastitis detection. These findings offer a scalable approach for enhancing herd health management, promoting economic sustainability, and enabling data-driven decision-making in resource-limited dairy systems.