Effective integration of vision foundational models for semantic segmentation to quantify grape foliage powdery mildew infection