Implementation of a Deep Learning System for Detection and Classification of Lumpy Skin Disease in Cattle: Enhancing...

Lumpy skin disease (LSD) poses significant threats to cattle health, mainly in the high bovine density locations, leading to severe agro-economic consequences. The present study is a maiden attempt to introduce deep learning (DL)-based systems to detect and classify LSD, leveraging two cutting-edge models: vision transformer (ViT) and ConvMixer. A comprehensive dataset of LSD lesions was developed, incorporating extensive data augmentation techniques for enhanced model robustness and generalization. The approach included the creation of an ensemble model that combined the strengths of both ViT and ConvMixer for improved accuracy of the diagnosis. The models were rigorously evaluated on the basis of the accuracy, recall, precision, F1-score and area under the receiver operating characteristic curve (AUC-ROC). The obtained results have demonstrated that the ensemble model outperformed the individual models significantly, achieving high precision and recall rates. The study showed the potentials of advanced DL techniques as reliable and efficient tools in veterinary diagnostics, offering early detection and management interventions to counter LSD outbreaks. Implementing these models could facilitate timely interventions to mitigate the economic impacts of LSD and improve livestock health management measures. This study opens up the scope to diagnose and classify numerous other neglected but critical bovine diseases by using similar digital platforms and employing new-age computerized tools such as artificial intelligence (AI) and machine learning (ML). To further enhance the practical applicability of DL-based diagnostic systems in future, investigations need to focus on expanding and enriching the datasets, refining model architectures and exploring real-time deployment. SUMMARY: Lumpy skin lesion images collected, and the LSD dataset is prepared for public use. Detecting and classifying lumpy skin lesions in cattle. Implementing DL models to classify lumpy skin lesions. Evaluated the efficacy of two select DL models using the balance or imbalance dataset. Measured performance of DL models on accuracy, precision, AUC, recall, F1-score, precision-recall curve and run time.