Transfer Learning-based Plant Disease Detection and Diagnosis System using Xception
Mohammed Abdul Moid, Mousmi Ajay Chaurasia
Abstract
Diseases in plants are one of the main threats to food safety. Some of the diseases in plants are infectious diseases that can spread throughout the entire field, thus affecting most of the yields. Therefore, early diagnosis of diseases in plants is necessary. For this, the proposed research work has developed a Transfer learning-based Plant disease classification system by using Xception architecture, which can correctly identify the diseases in plants and provide a solution to eliminate or prevent the spread of that disease when an image of the infected area is given as input. Apart from Xception architecture, this research work has also implemented the InceptionV3 model, upon which the former model is based on. These architectures were chosen due to their small size and low computational requirements. Using Xception architecture we achieved an accuracy of 97.5 percent. We have trained these models on a dataset containing 70,285 training images and 17562 validation images holding 38 different classes of healthy and diseased plants.