Wheat disease recognition: A Multi-Model Analysis with Customized CNN and Transfer Learning enhanced by Explainable AI techniques with LIME and Grad-CAM

The integration of deep neural learning and machine learning techniques into crop classification introduces an innovative dimension to agricultural progress. Concurrently, an image-based deep neural learning accession is employed to address the recognition and stratification of disease-affected wheat plants. The study’s core intent is to refine methods for diagnosing and categorizing wheat plant diseases, aligning with the broader aim of enhancing agricultural sustainability. The study unveils the potential of deep neural network models, prominently convolutional neural networks (CNNs), renowned for their effectiveness in image analysis tasks. A comprehensive assessment evaluates five distinct CNN models—VGG19, ResNet50, AlexNet, Inception V3, and DenseNet121—using a publicly accessible dataset for agricultural research referred to as the Large Wheat Disease Classification dataset (LWDCD2020) of 3814 wheat leaf images with four classes such as leaf rust, wheat loose smut, crown and root rot, which are fungal diseases and healthy plant images. The evaluation, focused on accuracy, reveals nuanced differences in performance. VGG19 emerges as the frontrunner with an impressive $\mathbf{9 6 \%}$ accuracy, followed closely by ResNet50, AlexNet, Inception V3, and DenseNet121. These nuances offer practitioners valuable insights for selecting optimal solutions. The study’s emphasis on transparency and clarity extends to the realm of explainable AI techniques. Prominent methods, Grad-CAM and LIME, are employed to shed light on the decision-making processes of deep neural networks (DNNs) models. In this context, Grad-CAM excels in providing more accurate and insightful explanations compared to LIME. These findings underscore the potential of deep neural networks (DNNs) in wheat disease classification and prediction. Beyond conventional performance metrics, the study underscores the significance of explainable AI methods in clarifying model predictions.