Disease Detection on Plant Leaves Using Deep Learning

Abstract

Diseases in plants present severe threats to food production across the world since they have a direct impact on agricultural production. In this study, deep learning solutions are sought to overcome the challenges of disease diagnosis on plant leaves. We collected a set of plant leaves pictures, and eight target variables defined plant species and disease states of concern. The training set contains 2589 images, while the test set contains 1300 images and each attribute has different numbers of samples. Deep learning algorithms of Xception, VGG16, ResNet50, InceptionV3, CNN01, and CNN02 were employed to train and test models for detecting diseases. As such, the results found in the experimental procedures portend accurate accuracy rates with CNN02 standing tall with a 98.10% accuracy. This shows how deep learning methods can work and the high accuracy in diagnosing plant diseases from images of leaves. Hence, this study provides valuable insights on potential developments for early disease diagnosis within an agricultural system while providing accurate and scalable solutions. As for the possible paths future research might follow, more emphasis might be placed on fine-tuning the Deep Learning models, as well as on integrating more types of sensors into the disease detection and management schemes within agriculture.