Classification of plant leaf diseaseImages using deep convolutionalNeural network

Abstract

The identification of plant leaf diseases is most significant toincrease crop yield and growth. Novel image-based plant leaf diseaseclassification models using the deep convolutional neural network arepresented in this research. Four data augmentation techniques are used toenhance the size and dimension of the plant leaf disease image dataset. Thedata augmentation techniques are Generative Adversarial Networks (GANs), newlineNeural Style Transfer (NST), four different color augmentation, and sevendifferent position augmentation techniques. These augmentation techniquesare increasing the performance of the dataset and balance the individual classsize of the dataset. The performance of the augmented datasets and theoriginal dataset was compared using the state-of-the-art transfer learningtechniques such as Inceptionv3, ResNet, and VGG16 networks.The four novel deep convolutional neural networks are a nine-layerdeep convolutional neural network, eleven-layer deep convolutional neuralnetwork, thirteen-layer deep convolutional neural network, and fifteen-layerdeep convolutional neural network. Theses deep convolutional neuralnetworks are developed using different sizes and numbers of convolutionaland pooling layers. Hyperparameters tuning techniques can find the value ofthe most suitable hyperparameter of any particular neural network. Randomsearch hyperparameters tuning technique was used to discover the value ofthe suitable hyperparameter to train for the four proposed models. Thegraphical processing unit is used to develop and train the nine-layer deepconvolutional neural network, eleven-layer deep convolutional neuralnetwork, thirteen-layer deep convolutional neural network, and fifteen-layerdeep convolutional neural network. newline newline