Towards Explainable Plant Pathology: Vision Transformer-Based Grape Leaf Disease Classification with LIME and SHAP
Ubiquitous Technology Journal · 18.11.2026 · 36-53
DOI: 10.71346/utj.v2i2.39 ↗Kurz gefasst
Fachartikel in der Zeitschrift Ubiquitous Technology Journal, online erschienen am 18.11.2026. Verfasst von Md. Rifat Hossen und 2 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY-SA 4.0).
Abstract (Originalsprache)
Grapevine diseases pose a significant threat to global viticulture, affecting both yield and quality, and necessitating the timely and accurate detection of these diseases. Traditional manual inspections are labour-intensive, subjective, and can lead to misdiagnosis. This study presents a reliable grape leaf disease (GDL) classification system that integrates deep learning techniques with explainable artificial intelligence (XAI), thereby enhancing detection accuracy and transparency. A publicly accessible grape leaf dataset, covering four categories —Bacterial Rot, Downy Mildew, Powdery Mildew, and Healthy Leaves —was preprocessed by resizing, normalizing, filtering, and applying extensive augmentation to improve model robustness. Several deep learning models, such as VGG19, ResNet50, Xception, and Inception, were evaluated alongside a Vision Transformer (ViT)-based model. The ViT demonstrated top performance, achieving 99% accuracy, 99% precision, 97% sensitivity, and 99% specificity, surpassing traditional CNNs and ensemble models. Additionally, feature selection techniques such as PCA and LDA enhanced class separation, while ensemble classifiers increased robustness. To enhance interpretability, LIME and SHAP were used to identify disease-relevant regions, promoting model transparency and trustworthiness. The comparison with existing approaches shows that our proposed method is superior. Overall, this research offers an efficient, interpretable, and scalable solution for automated grape disease detection, aiding smart agriculture and sustainable crop management.
Zitieren
APA
Hossen, M. R., Al-Faruk, J., & Sarkar, S. (2026). Towards Explainable Plant Pathology: Vision Transformer-Based Grape Leaf Disease Classification with LIME and SHAP. Ubiquitous Technology Journal, 36-53. https://doi.org/10.71346/utj.v2i2.39
Vancouver
Hossen MR, Al-Faruk J, Sarkar S. Towards Explainable Plant Pathology: Vision Transformer-Based Grape Leaf Disease Classification with LIME and SHAP. Ubiquitous Technology Journal. 2026:36-53. doi:10.71346/utj.v2i2.39
BibTeX
@article{Hossen2026,
author = {Hossen, Md. Rifat and Al-Faruk, Jonayed and Sarkar, Sakiba},
title = {Towards Explainable Plant Pathology: Vision Transformer-Based Grape Leaf Disease Classification with LIME and SHAP},
journal = {Ubiquitous Technology Journal},
year = {2026},
pages = {36--53},
doi = {10.71346/utj.v2i2.39}
}
EUVIMED