Comparative evaluation of CNN architectures for wheat rust diseases classification

Wheat is a critical staple crop that plays a key role in global food security, particularly in developing countries like Ethiopia. However, fluctuations in wheat yields, often driven by factors like disease, pose significant challenges to agricultural productivity in these regions. Brown rust, stem rust, and yellow rust are the major types of rust diseases that generally cause massive reductions in crop quality and yield. An automated deep learning technique is needed for the classification of these diseases in agricultural fields. In this study, different CNN Architectures were compared to classify wheat rust diseases from wheat image data. The data is split into training, testing, and validation with ratios of 65:20:15, respectively. MobileNetV3, DenseNet121, ResNet50, and EfficientNetB0 were utilized to classify wheat rust disease. Among all of these, MobileNetV3 outperformed these models with an accuracy of 97.7%. Hence, due to this performance, MobileNetV3 became the best model for wheat crop disease classification and management.