Combined CNN with STARGAN for Wheat Yellow Rust Disease Classification

Abstract

Experienced evaluators takes a lot of time for the correct prediction of wheat yellow rust. For better prediction of wheat yellow rust disease in wheat plant, computer assisted techniques such as machine learning (ML), deep learning (DL), image processing and computer vision techniques are employed. While considering these aspects, an automatic classification system for wheat yellow rust using a hybrid approach of a generative adversarial network (GAN) and convolutional neural network (CNN) is proposed. Also, the proposed model helps to identify wheat yellow rust disease at different severity levels (very low, low, medium, high, and very high). For achieving objectives about proposed model, State of Art GAN (STARGAN) helps in data augmentation of wheat plant disease images. Finally, a wheat rust model based on convolutional networks was trained on generated data. A case study has also been accomplished on the generated data by deploying a CNN model for classification. Then a comparative analysis of the proposed methodology has been consummated by differentiating the performance metric with the fully convolutional network, deep CNN (DCNN) and random forest models. A Deep CNN is a CNN model with a huge number of hidden layers. Comparing the proposed approach with other models, the proposed approach achieves a greater classification accuracy of 95.6% at a medium severity level.