An implementation and analysis of deep learning models for the detection of wheat rust disease
Shivani Sood, Harjeet Singh
Abstract
Wheat is one of the essential food crops all over the world. In India, it is the largest growing winter cereal crop that contributes almost 14% to the world's food production. This crop is usually affected by various types of rust diseases. Rust is one kind of fungal disease, which occurred on wheat crop. The rust is characterized as leaf-rust (brown rust), stem-rust (black rust), and yellow rust (strip rust). Due to these types of diseases, a huge amount of loss may occur in the production of wheat crop. Therefore, it is required to early detect this disease during the growing stage. In the present study, we consider two classes of diseased plants and one for a healthy plant. Due to less amount of image dataset, we performed the image augmentation and generated the synthetic data from the original data. Various experiments are performed on different dataset sample sizes i.e., 100, 500, 1000, 1500, and 2000. The weights for building the model have been considered as the same in ResNet50 and VGG16 transfer learning models. Moreover, to improve the accuracy of the model, we tune the hyper-parameters. The hyper-parameters used to train this model are: batch_size = 32, optimizer = "adam", initial learning rate = 0.01 with decay of 0.0001. After performing the experiments, the model produced appreciable results with the VGG16 model. The proposed approach has obtained the highest classification accuracy rate of 99.07% using the VGG16 model.