Adams Algorithm with Generative Adversial Network for Maize Leaf Disease Classification

Abstract

The diseases in plant revealed overwhelming bang on providing security in food production and may cause diminution in quality of agriculture-assisted items. In several scenarios, the disease in plant led to no grain produce. Hence, automated detection of plant disease is suggested for discovering the information regarding agriculture. Various methods are developed to discover plant disease wherein deep model is mostly favoured. A new model is devised for maize leaf disease classification. Here, the pre-processing is a preliminary step to eradicate noise contained in image and implemented with ROI extraction and anisotropic filter. Thereafter, the saliency map extraction is adapted for computing the quality of each pixel. The purpose of saliency map extraction is to alter representation of image into something which is more meaningful or easy to inspect. Finally, the classification of maize leaf disease is executed with Generative Adversial network (GAN), and helps to classify whether the maize leaf is normal or infectious. The GAN training is implemented with Adam. The Adam GAN showed best performance with improved accuracy of 0.811, sensitivity of 0.879 and specificity of 0.939.