SmartFarming: Lumpy Skin Disease Detection from Smartphone-captured Cattle Images using Transformer Model

Lumpy skin disease, commonly known as LSD, poses a significant threat to cattle health, welfare and the livestock industry. It is associated with substantial economic loss and cattle death, especially in Bangladesh, where most farmers do not have proper training. In 2023, the marginal farmers of Bangladesh experienced a massive loss due to the LSD outbreak. Existing methods for detecting LSD include physical examination and laboratory tests such as polymerase chain reaction or microscopic observation. Physical examination to detect LSD at an early stage requires training and laboratory tests are expensive and time-consuming. This highlights the need for a practical, cost-effective, and rapid system for LSD detection. Artificial intelligence (AI) enabled computer vision can facilitate LSD detection. Several methods previously proposed utilizing this technology failed to achieve adequate accuracy and practical usability. This paper presents an AI-guided system that enables early LSD detection from smartphone-captured low-resolution images. This system utilized a vision transformer (ViT) model to detect LSD, which achieved 99% test accuracy and 97% 5-fold cross-validation accuracy when demonstrated on the heterogeneous dataset. It allows on-site LSD detection in less than a minute without requiring any prior experience. High accuracy, rapid speed and robustness ensured the practical use of the system.