AI-Driven Remote-Sensing Framework for Early Detection of Wheat Rust Using UAV Multispectral Imagery and Leaf Spectral Signatures

Wheat rust is one of the most destructive fungal diseases affecting wheat production and food security, especially in regions where timely field monitoring is limited. Conventional disease scouting is labor-intensive and often identifies infection after visible symptoms appear. This study presents an AI-driven remote-sensing framework for early wheat rust detection using UAV-based multispectral imagery, leaf-level spectral signatures, vegetation indices, machine learning, and mobile decision support. The study was conducted at the National Agricultural Research Centre (NARC), Islamabad, Pakistan, during the active wheat-growing season. UAV multispectral imagery was acquired using five spectral bands: Blue, Green, Red, Red Edge, and Near-Infrared. Leaf-level spectral measurements were also collected using ASD FieldSpec data in SED format to capture plant-level reflectance variations associated with healthy and diseased wheat conditions. The UAV data were preprocessed through radiometric correction, filtering, normalization, geometric correction, and band co-registration. Vegetation indices, including NDVI, GNDVI, NDRE, LCI, and NDWI, were computed to represent vegetation vigor, chlorophyll variation, Red-Edge sensitivity, and moisture-related stress. Spectral, vegetation index, spatial, and texture features were extracted and used to classify healthy and diseased wheat samples. Classical machine learning models, including Random Forest, Support Vector Machine, Artificial Neural Network, and XGBoost, were compared for classification performance, while U-Net was considered for pixel-level disease mapping. The results showed that healthy wheat areas generally produced stronger NIR response and higher vegetation index values, whereas diseased or stressed regions showed reduced spectral response. Red-Edge-based features, particularly NDRE and LCI, were useful for early disease stress detection. Overall, this study demonstrates the potential of integrating remote sensing and AI for timely, data-driven wheat disease management.