SEVERITY CLASSIFICATION OF YELLOW RUST DISEASE AND TREATMENT RECOMMENDATION FROM WHEAT IMAGES USING DEEP LEARNING TECHNIQUES

Abstract

Wheat yellow rust is the most prevalent and severe wheat fungal disease caused by Puccinia striiformis f. sp. tritici, a pathogen that causes significant losses in wheat production globally because of its highly destructive properties. The existing manual identification and diagnosis practice is very expensive and time-consuming, leading to errors. Therefore, automatic identification of these diseases through a system is critical. Thus, the objective of this study is to develop an automated system that can classify and provide treatment recommendations for wheat yellow rust. Design science research methodology was used to achieve this objective. Various studies conducted in this area have not identified the severity levels or provided treatment recommendations for wheat yellow rust. During problem identification, the researcher attempted to obtain valuable information on the selected diseases from the domain expert through interviews, observation, and document analysis. A total of 10,000 images were gathered and prepared from the Kulumsa Agricultural Research Center and farming areas. In this study, image preprocessing techniques such as normalization, noise removal, filtering, and segmentation were used for region of interest extraction, a neural network for feature extraction, and SoftMax as a classifier. The dataset was split into 70% training, 15% validation, and 15% testing datasets. For experimentation, both the Proposed and CNN pretrained models (InceptionV3, InceptionResnetV2, MobileNetV2, and Combined) were used, and their performances were compared using accuracy, precision, recall, and f1-score performance measurement methods. Given the wheat leaf image, the proposed Wheat Rust Network (WRNet) model classifies as healthy, resistant, moderately resistant, moderately susceptible, or susceptible according to the rust severity with treatment recommendation in percent. The proposed model achieved 99.11% training accuracy, 99.04% validation accuracy, and 99% testing accuracy for grading and treatment recommendations for wheat yellow rust. The model proposed in this study improves the performance of state-of-the art CNN models. Further studies are needed to extend the proposed model to the detection and classification of other wheat diseases.