Optimized Deep Convolutional Generative Adversarial Network for Tomato Leaf Disease Identification

Abstract

The yield of tomato are mainly affected by leaf disease, which can be accurately identified from the leaf images using deeplearning techniques. Already available deep learning techniques use data augmentation methods to increase the dataset size, that comprise translation, rotation and flip. But it does not result in good generalization. The aim is to attain better accuracy and to achieve good generalization results. Instead of normal data augmentation procedures Deep Convolutional Generative Adversarial Networks (DCGAN) is proposed here. Initially Image classification will be done using different CNN architectures such as VGG16, InceptionV3 and ResNet50 models with the original dataset and then the results are compared with models trained by augmented datasets using GAN. The models trained with datasets augmented by GAN provided good accuracy.