AI-Based Crop Disease Severity Assessment with Explainable AI: A Systematic Review of Methods, Trends, and Research Gaps

Crop diseases have a major impact on agricultural productivity and food security. Therefore, the timely detection and accurate severity assessment of diseases are of importance to effective crop management. Automatic crop disease detection and classification have been extensively implemented using artificial intelligence (AI), especially deep learning and computer vision. However, direct evaluation of disease severity and trustworthiness of model explanations are still important aspects to be explored. The review discusses recent progress on AI based assessment of crop disease severity and Explainable Artificial Intelligence (XAI . Selected studies are analyzed based on type of crop and disease, datasets, AI models, severity evaluation methods, XAI methods, reported performance, limitations and gaps in research. The review explores various approaches such as convolutional neural networks, attention-based architectures, transformer-based models, segmentation techniques, hybrid frameworks, and few-shot learning. Approaches to severity assessment range from assigning diseases to predefined severity categories to quantifying disease severity and forecasting disease progression or related crop damage. We consider the utility of XAI methods such as Grad-CAM, Grad-CAM++, SHAP, LIME, and Integrated Gradients for interpreting model predictions and for identifying disease relevant regions or features. Furthermore, the review covers supporting studies regarding crop disease detection, classification, and risk prediction to comprehend their role and limitations for severity assessment. Several challenges are highlighted in the analysis including scarcity of standardized severity labeled datasets, inconsistent severity grading methods, limited crop and geographical scope, inadequate real-field validation, limited deployment of lightweight models and insufficient evaluation of explanation reliability. The findings highlight the need for robust AI models that can accurately assess disease severity and provide meaningful explanations across different agricultural settings. his review summarizes recent progress in AI- and XAI-based crop disease assessment, identifies important research gaps, and suggests future research directions toward the goal of building reliable, interpretable, and practical systems for precision agriculture.