Abstract
Yellow rust, caused by Puccinia striiformis f. sp. tritici, represents one of the most severe biotic threats to winter wheat (Triticum aestivum L.) production across major agricultural belts in Eurasia. Timely and accurate detection of infection during early development stages is critical for minimizing yield losses and optimizing site-specific fungicide applications. In this study, an integrated framework combining high-resolution spectral and spatial feature extraction from Unmanned Aerial Vehicle (UAV) hyperspectral imagery (400–1000 nm) was evaluated for early-stage yellow rust identification in winter wheat micro-plots. Field trials were established under controlled inoculation and natural infection regimes. Spectral reflectance indices responsive to chlorophyll degradation, carotenoid accumulation, and leaf structure changes were extracted alongside texture parameters derived from Gray-Level Co-occurrence Matrix (GLCM) analysis. Machine learning classifiers, including Random Forest (RF) and Support Vector Machines (SVM), were trained using standalone spectral, standalone spatial, and integrated spatial-spectral feature sets. The integrated model achieved a peak overall accuracy of 92.4% and a Cohen's kappa coefficient of 0.89 during the early asymptomatic-to-incipient symptomatic stage, significantly outperforming single-domain models by 8.1% to 11.5%. Spectral regions in the red-edge (705–725 nm) and near-infrared (750–850 nm) combined with GLCM texture entropy and contrast proved to be the most sensitive indicators of early pathogenesis. These findings demonstrate that fusing spectral and spatial context from UAV hyperspectral platforms provides a robust tool for early fungal disease monitoring in precision cereal production systems.
