A Comparative Analysis for Grapes Leaf Disease Diagnosis Using Deep Learning Models

Grape leaf diseases significantly impact vineyard productivity and grape quality, necessitating efficient and accurate diagnosis methods. Unfortunately, when performing manual inspection, it can be inconclusive and extremely time- consuming which underlines the importance of an automated approach. In this paper, a comparative assessment of four deep learning models, DenseNet, MobileNet, EfficientNet B7, and AlexNet has been conducted for the diagnosis of the grape leaf disease. The performance of each model is assessed by accuracy, precision, recall, and, F1-score that is relevant to its applicability in the practical agriculture-related environment. The chief innovation of this work is therefore based on main open issues as imbalanced datasets and variations on the environment that can hinder the reliability of current models. The problem of model and feature resilience is addressed by applying data augmentation techniques, including GANs, to ensure that the models can generalize from one vineyard condition to another. The findings also show that the approach that uses EfficientNet B7, DenseNet, and transfer learning & ensemble method can provide higher accuracy and robustness for using in vineyard. The research presented here benefits from the progress of smart agriculture in providing a practical and effective solution for distinguishing grape leaf diseases.