Auto encoder Based Tomato Leaf Disease Identification

Abstract

The yield of tomatoes mainly affects leaf disease. It can be detected accurately using advanced deep learning techniques. Initially Autoencoder is used to remove the noise as an initial preprocessing step. Various autoencoders like Simple autoencoder based on fully-connected layer, sparse autoencoder, deep fully connected autoencoder and image denoising autoencoder are used and compared. As a second step Deep Convolutional Generative Adversarial Network(DCGAN)is used for augmentation purpose instead of traditional augmentation method such as translation, rotation and flip. Since the traditional technique does not result in good generalization DCGAN is used to attain better accuracy and to achieve good generalization results. Finally diseases are classified using VGG16 architecture. We also made a comparison with the results to find out the difference between using with and without autoencoder along with hyperparameter tuning. We provided a detailed explanation of how these algorithms work and comparison between them.