Multi-Class Classification and Transfer Learning for Rapid Disease Detection in Green Leafy Vegetables Using Convolutional Neural Networks

Abstract

Plant diseases cause various pathogens, including bacteria, viruses, fungi, and protozoa. Plant diseases and pathogens have an impact on the entire crop and field area. Plant diseases can be identified by various techniques; however, traditional techniques are time-consuming and require more effort. Up-to-date machine learning based approaches are used to identify plant diseases based on the image data. This study investigated the application of convolutional neural network (CNN) models for the detection and classification of diseases affecting green leafy vegetables. Several state-of-the-art CNN architectures, including InceptionV3 and DenseNet121, were evaluated on an image dataset of diseased and healthy leaf samples. Preprocessing techniques such as scaling, cropping, grayscale conversion, and normalization were applied to enhance the input images. The CNN models demonstrated high diagnostic accuracy, with InceptionV3 and DenseNet121 exhibiting exceptional performance across multiple metrics like sensitivity, specificity, accuracy, recall, F-measure, and Matthews correlation coefficient (MCC).