Plant Leaf Disease Classification using Local-Based Image Augmentation and Convolutional Neural Network

Abstract

A local-based image augmentation method to cooperate with convolutional neural networks for plant leaf disease classification is proposed. There are three processes of the diseased classification such as: 1) the local-based image augmentation method, 2) the global-based image augmentation methods and 3) convolutional neural network (CNN) models. First process, the local-based image augmentation method is designed to distribute disease regions on the plant leaves for increasing the number of images in the datasets. The proposed method focuses on only the regions on the leaves which is called the local-based approach. Second process, the general global-based image augmentation methods are used to increase variety of the images regarding pose, brightness, blur and noise. Third process, the simple CNN model is proposed for plant leaf disease classification using the augmented images. The CNN model is proposed in this research. The VGG19 and MobileNet are using for comparison and study the impact of the proposed method in the first process including. We test on two kinds of plants: potato and grape with three CNN models: our CNN model, VGG19 and MobileNet. The accuracy rate of potato and grape leaf for our CNN, VGG19 and MobileNet model are 93.78%, 95.56%, 77.33%, 93.90%, 96.30%, and 80.00%, respectively. From the result can be concluded, that the proposed local-based image augmentation method can increase the accuracy rates of all models.