Thermal Image-Based Artificial Neural Network Approach to Determine Mastitis Detection in Holstein Dairy Cattle

Mastitis, a disease associated with milk production with multiple etiologies, causes significant economic losses among dairy farmers worldwide. This study aimed to detect mastitis using thermal images of the udder obtained during the milking phase from 500 Holstein dairy cows with the aid of an Artificial Neural Network (ANN). Mastitis levels were classified based on the California Mastitis Test (CMT) scores using somatic cell count (SCC) as the output variable. The dataset was divided into training (70%), validation (15%), and test (15%) subsets. RGB (Red, Green, Blue) thermal images were used to construct the input matrices. The model achieved correlation coefficients (R) of 0.91, 0.97, and 0.97 for the training, validation, and test datasets, respectively. The close agreement between validation and test performances indicates the absence of overfitting and demonstrates strong generalization capability of the proposed model. These findings suggest that artificial neural networks combined with thermal imaging can provide high-quality and reliable results for mastitis detection.