Enhancing Grapevine Leaf Classification Through Deep Learning with VGG 19 Architecture
Goldy Verma
Abstract
In this research paper, the categorisation of images of the grapevine leaf by the application of deep learning strategies by the fine-tuned VGG19 model is traversed. Grapevine leaves are classified into five well-defined classes: Ak, Ala_Idris, Buzgulu, Dimnit and Nazli. The detailed collection of labelled grapevine leaf images is extracted from the Kaggle open-source platform. For escalating the model execution, we employed preprocessing steps, including resizing, normalisation, and augmentation of images. The fine-tuned VGG19 model is made to order with 5 convolutional layers and 4 max pool layers for encountering the complicated leaf features effectively. The model was well generalised and therefore, validates using an 80:10:10 split of the dataset, it involved the Adam optimiser and a loss function which measures the distinction between the actual and predicted categories. The model has depicted its capability to classify the grapevine leaf images accurately by achieving an accuracy of 80%. The detailed analysis through loss and accuracy plots and the confusion matrix highlights the robust performance and the areas of potential refinement. This research highlights the efficiency of utilising deep learning models like fine-tuned VGG19 for the automation of grapevine diagnosis, promoting a reliable for precise grapevine leaf classification.