Interpretable disease classification in plant leaves using deep convolutional neural networks

Abstract

Agriculture has been crucial for centuries. Due to its revenue contribution, agriculture's importance has grown throughout time. However, some counter factors prohibit us from getting the full benefits of crops. Natural plant diseases are one factor. The main causes of these difficulties are harsh weather and excessive pesticide use, which strain Bangladesh's economy. To lessen the problem's severity, an image processing system was created that uses Deep Learning and CNN to classify leaf illnesses. The primary demographic is farmers and others who are willing to tend crops. It was decided to make sure the proposed model is lightweight so that it can be compatible and simple to implement on low-end devices without using up excessive resources. This CNN algorithm predicts the leaf's status based on the user's selected images. After constructing CNN, another model is offered, LIME, based on Explainable AI (XAI). XAI helps humans understand AI's decisions or predictions. After the proposed CNN model diagnoses diseased leaves, the XAI helps us understand why. Conclusively, 99.87%, 99.54%, 99.54% accuracy was found in training, validation and testing respectively after running our models.