Hybrid Deep Learning Approach for Early Mastitis Detection in Dairy Cattle Using Convolutional Neural Networks and Vision Transformers

Mastitis poses a major health challenge in dairy cattle, resulting in considerable economic losses due to reduced milk production and quality. Early detection is crucial for effective management and prevention of the disease. Traditional diagnostic methods, often relying on physical examination and laboratory tests, are insufficient for early-stage detection. This paper introduces a novel approach to mastitis detection by combining Convolutional Neural Networks (CNNs) with a Vision Transformer (ViT) model, leveraging the strengths of both architectures. In this study, a hybrid model is developed, where a pre-trained CNN such as ResNet50 is utilized to extract spatial features from udder images. These features are then processed by the Vision Transformer (ViT), which excels in capturing complex relationships and dependencies within the image data. The self-attention mechanism in ViT enables it to efficiently process extracted features, enhancing detection accuracy. The hybrid model is refined using a diverse dataset of udder images, including both healthy and mastitis cases. The results demonstrate that the integration of the Vision Transformer with a CNN significantly enhances the model's ability to detect mastitis at an early stage, offering a promising solution for automated and accurate diagnosis. This approach has the potential to revolutionize disease management practices in the dairy industry, leading to better outcomes for farmers and livestock.