Deep Learning-Based Grapevine Leaf Classification Using a Fine-Tuned VGG19 Architecture

This study work traverses the classification of photographs of the grapevine leaf by means of deep learning algorithms applied under the fine-tuned VGG19 model. There are five well established classifications of grapevine leaves: Ak, Ala_Idris, Buzgulu, Dimnit and Nazli. The Kaggle open-source platform provides the comprehensive collection of tagged grapevine leaf photos. We used preprocessing techniques like downsizing, normalising, and picture augmentation to increase the model execution. With five convolutional layers and four max pool layers to efficiently address the complex leaf characteristics, the refined VGG19 model is built. The model was effectively generalised hence it validates with an 80:10:10 split of the data. It comprised the Adam optimiser and a loss function measuring the variations between the predicted and actual groups. Reaching an accuracy of $80 \%$, the model has demonstrated its ability to exactly classify the images of grapevine leaves. The detailed study through loss and accuracy graphs and the intricate confusion matrix emphasizes both the complete performance and the areas of prospective improvement. This work emphasizes the effectiveness of using deep learning models, including finely calibrated VGG19, for the automation of grapevine diagnosis, so allowing a consistent solution for exact grapevine leaf classifying. Resilient agricultural systems and sustainable development are supported by this effort in line with sustainable farming practices, innovation in food security, climate adaptation, resource economy, and environmental protection.