Wheat Leaf Disease Detection Using Shallow ResNet50V2

Plant diseases are primary concerns of farmers globally. Significant yield losses directly affect the economy. Wheat is considered one of the major foods, affected by viruses, bacteria, and fungi that cause productivity. Agronomists are unavailable in rural areas to identify these diseases from this raw data. Earlier detection techniques mainly depend on visual observation of plant leaves which often misdiagnose the actual disease and as a result wrong medicines are imposed on the plants. Image data can now be processed through machine learning-based techniques, which are cost-efficient and perform better than previous methods. The popular machine learning-based technique named Convolutional Neural Network (CNN) has shown remarkable capabilities in effectively processing these raw input data through multiple layers of convolution, pooling, and dense. However, training of these networks demands a substantial amount of data and results in extensive training time and memory resources. Additionally, they frequently encounter overfitting problems, leading to reduced recognition accuracy. To address this problem, we propose a shallow network. Due to the ability to handle vanishing gradient problems using skip connection, we considered ResNet50V2 architecture and designed a shallow version of this framework. Through our proposed system, we were able to achieve an impressive accuracy rate of 97.00%, which serves as strong evidence supporting the effectiveness and success of our approach in tackling the challenges associated with detecting and addressing these diseases on wheat plants.