Deep Learning-Based Classification of Grape Leaf Diseases using EfficientNetB0 and InceptionV3 for Enhanced Vineyard Management

This work proposes an effective deep learning method for the classification of grape leaf diseases, which aims to contribute to early detection and management in vineyards. This implementation of transfer learning used two pre-trained models, EfficientNetB0 and InceptionV3, both initiated on ImageNet weights and fine-tuned on a four-class categorization dataset consisting of images representing Black Rot, ESCA, Healthy, and Leaf Blight manifestations in grape leaves. Data augmentation techniques included random flipping and zooming to improve model generalization. Categorical crossentropy loss and the Adam optimizer were set during training. These models are evaluated on accuracy, precision, recall, and F1-score metrics. EfficientNetB0 outperforms InceptionV3 by a margin, with an overall accuracy of $\mathbf{9 8 \%}$, while InceptionV3 achieved 92% accuracy. EfficientNetB0 had a better performance in precision, recall, and F1-score for all classes than others. That would actually indicate its higher performance in detecting grape diseases. By pointing out the potential of EfficientNetB0-which can give reliable classification-the tool for vineyard disease management would be appropriate. This is a research study that has contributed toward advances in the field of auto-detection of a disease by presenting an accurate, efficient, and scalable solution which helps boost vineyard productivity through early intervention with effective disease management strategies.