Deep Learning based Grapevine Leaf Classification using Augmented Images and Multi-Classifier Fusion for Improved Accuracy and Precision

Grapevines mainly produce grapes, which can be enjoyed fresh or processed. Grapevine leaves are also collected annually as a secondary product, with their variety impacting both pricing and flavor. This research utilizes Deep Learning techniques to classify grapevine leaves based on images. Initially, 500 images of vine leaves representing five species are captured using a unique self-illumination system. Data augmentation methods expand the dataset to 2500 images. Classification is conducted using a modern CNN model, specifically fine-tuned MobileNetv2. Data preparation involves training the dataset with three classifiers: basic CNN, VGG19, and MobileNet. Image processing converts the RGBA dataset into RGB format. The dataset is split into three subsets: 70% for training, 20% for validation, and 10% for testing. Model preparation offers three options for selecting the appropriate architecture: Simple CNN, VGG19, and Transfer Learning with MobileNet. Model analysis assesses performance using key metrics and visualizations, achieving an accuracy of 96% with Adam optimizer and 88% with Stochastic Gradient Descent.