GrapeVineGuardAI: A Robust Deep Learning Framework for Automated Grapevine Leaf Disease Recognition Using Vision Transformers and Adaptive Multi-Level Feature Aggregation

Abstract

Grape cultivation plays a significant role in global agriculture; however, grape leaf diseases such as Black Rot, Esca (Black Measles), and Leaf Blight pose major challenges to crop productivity and quality. Conventional disease diagnosis relies on manual visual inspection, which is labour-intensive, time-consuming, and often inaccurate for large-scale vineyard monitoring. To address these limitations, this study proposes an automated grape leaf disease detection framework based on deep learning and computer vision techniques. A dataset comprising 7,222 grape leaf images belonging to four classes—Healthy, Black Rot, Esca, and Leaf Blight—was utilized for model development. The images were preprocessed using resizing (224 × 224 pixels), normalization, and data augmentation techniques to improve model robustness and generalization. Three deep learning architectures, namely Convolutional Neural Network (CNN), DenseNet, and EfficientNet, were implemented and comparatively evaluated for disease classification. Experimental results demonstrated that the EfficientNet model achieved the highest classification performance with a testing accuracy of 99.03%, outperforming the CNN (98.06%) and DenseNet (92.52%) models. The trained model was integrated into a Flask-based web application, enabling users to upload grape leaf images and obtain real-time disease predictions with confidence scores. The proposed framework provides a rapid, accurate, and scalable solution for early grape leaf disease diagnosis, facilitating timely intervention and reducing crop losses. The obtained results demonstrate the effectiveness of EfficientNet for intelligent plant disease classification and highlight its potential application in precision agriculture and smart farming systems.