Advancing Grapevine Leaf Classification Through Transfer Learning: A Fine-Tuned VGG19 Deep Learning Approach
Rajesh Dey, Deepak Banerjee, Goldy verma, Manish Kumar Singla, Ashish Kumar Singh, P William, Rupali Atul Mahajan, Monish Khan
Abstract
The classification of images of the grapevine leaf through the use of deep learning methods by the finetuned VGG19 model is traversed in the study. There are five distinct classes of grapevine leaves: Ak, Ala idris, Buzgulu, Dimnit and Nazli. The labeled collection of image of grapevine leaves, in detail, is harvested in the Kaggle opensource platform. To scale up the model implementation, we used preprocessing methods, such as resizing, normalisation, and image augmentation. The fine-tuned VGG19 model is custom-designed to tackle the complex leaf structures with 5 convolutional layers and 4 max pool layers. The model was highly generalised and thus it can be validated with an 80:10:10 split of the data, it used the Adam optimiser and a loss that quantifies the difference between the actual and predicted categories. The model has portrayed its ability to categorize the images of the grapevine leaves with accuracy of 80%. The loss and accuracy plot analysis and the complex confusion matrix present the results of the strong performance and the aspects that can be improved. This study underscores the effectiveness of using deep-learning-based models, such as a fine-tuned VGG19, in the automation of grapevine diagnosis to support a trustworthy approach to grapevine leaf classification.