Distinguishing lumpy skin disease from coat patterns using morphological priors in deep learning.

Lumpy skin disease (LSD) is an important transboundary disease of cattle that compromises animal welfare and causes major production and economic losses. Early on-farm recognition is challenging, particularly when lesions are small, low contrast, or obscured by complex coat patterns. Meanwhile, the growing use of on-farm cameras and smartphones creates opportunities for image-based tools to support herd-level screening and earlier veterinary intervention. Here, we develop a morphology‑driven image analysis model for lumpy skin disease detection in cattle (LSDD). The framework is designed to reflect visual cues used by veterinarians: a local texture refinement module highlights subtle nodules while suppressing background noise, and a global morphology consistency module emphasizes the circular or clustered configurations typical of true lesions while down-weighting elongated coat markings. The system outputs an image-level classification ("healthy" vs "lesion") and is intended to complement-not replace-clinical examination and laboratory testing. Using datasets that reflect typical farm conditions, we conducted extensive experiments under varying illumination, backgrounds, and coat patterns. Overall, our results suggest that LSDD is a practical morphology-driven screening aid for herd-level, image-based monitoring, helping to flag animals with LSD-compatible lesions earlier and thereby strengthening on-farm surveillance and supporting timely implementation of control measures.