Advanced Deep Learning Approaches: Utilizing VGG16, VGG19, and ResNet Architectures for Enhanced Grapevine Disease Detection

• Main Authors: Yadav, Ajay Pratap Singh, Thapliyal, Nitin, Aeri, Manisha, Kukreja, Vinay, Sharma, Rishabh
• Format: Conference Proceeding
• Language: English
• Subjects: Agricultural AI; Biological system modeling; Convolutional neural networks; Crops; Deep learning; Grapevine Disease Classification; Pipelines; Plant Pathology; Soil; VGG16; VGG19; Visualization
• ISSN: 2769-2884
• DOI: 10.1109/ICRITO61523.2024.10522276

This study aims to give an overall comparison among three of the most powerful convolutional neural network (CNN) architectures-VGG16, VGG19, and ResNet for the classification of leaf diseases, particularly targeting leaf blight, ESCA, black rot, and healthy leaves. With the consequent economic effects of these diseases on viticulture, it is essential to create precise, productive, and scalable diagnostic tools for the pursuit of sustainable agricultural practices. Using a very large dataset of grapevine leaf images acquired from different localities, this study investigates the ability of the models to correctly assign the diseases to the visual symptoms. The models were assessed through a rigorous evaluation process including data preprocessing, augmentation, and stratified k-fold cross-validation approach to determine the overall accuracy, precision, recall, and F1 score. The study results reflect no doubt that the ResNet model surpasses its rivals with an outstanding accuracy of 95% which is followed by VGG19 with 93.5% and VGG16 with 92%. These observations emphasize the power of deep residual learning and the importance of architecture depth in ensuring the accuracy of CNNs in the field of plant disease classification tasks. In addition, the study underlines the need for having holistic and varied datasets that train models to generalize across different environments and disease presentations. The current research adds to the growing knowledge base on the use of deep learning (DL) in agriculture, thus providing insight into the most efficient CNN networks that can be used for grapevine disease recognition. This work contributes to the integration of AI-based diagnostic tools in precision agriculture; this will eventually help improve the current disease management practices and foster sustainable crop production systems.