Abstract
Wheat rust is a severe fungal disease that significantly impacts wheat crops, resulting in substantial losses in quality and quantity, often exceeding 50%. Timely and early accurate estimation of disease severity in fields is critical for effective disease management. Early identification of Rust at low severity levels can facilitate prompt implementation of control measures, potentially saving crops. This paper introduces an automated wheat rust severity stage estimation model utilizing the EfficientNet architecture and attention mechanism. The convolutional Block Attention Module was integrated into EfficientNet-B0 in place of the SE module to enhance feature extraction by simultaneously considering channel and spatial information. The proposed hybrid approach aims to identify rust disease severity accurately. The model is trained on an image dataset comprising three major rust types-stripe, stem, leaf, and healthy plants captured under real-life field conditions. Each disease is categorized into four severity stages: healthy, low, medium, and high. Experimental results demonstrate that the proposed model achieves impressive performance, with a training accuracy of 99.51% and a testing accuracy of 96.68%. Moreover, comparative analysis against state-of-the-art CNN models highlights the superior performance of our approach. An Android application was also designed and developed to facilitate real-time classification of plant disease severity. This system incorporates a severity model optimized for enhanced classification accuracy and rapid recognition, ensuring efficient performance.