An early detection model of wheat stripe rust utilizing two-dimensional correlation spectroscopy for eliminate growth-related interference
Liang, Xiaoying ; Jiang, Guihe ; Shi, Huanrui ; Mao, Yunfei ; Wang, Ning ; Wang, Xiaojie ; Chen, Xu
Wheat stripe rust, caused by Puccinia striiformis f. sp. Tritici (Pst), represents a significant threat to global wheat production. Early detection, particularly during the asymptomatic phase, is critical for effective disease management. Hyperspectral sensing can detect subtle physiological alterations associated with initial infection; However, its effectiveness is frequently limited by substantial background interference from normal plant growth. In this study, the reliability of hyperspectral data obtained from early asymptomatic leaves was first validated using quantitative real-time polymerase chain reaction (qPCR). To mitigate background the interference, generalized two-dimensional correlation spectroscopy (2D-COS) was employed, utilizing infection time as the perturbation variable. This approach surpasses conventional dimensionality reduction techniques such as principal component analysis (PCA) and the chemometric feature selection algorithm known as competitive adaptive reweighted sampling (CARS). Through this methodology, six feature bands exhibiting distinct absorption changes were identified. Analysis of synchronous and asynchronous 2D-COS correlation features from 1 to 6 days post-inoculation (dpi), enabled effective discrimination between spectral variations attributable to growth and those specific to disease responses. The biological significance of these spectral dynamics was empirically validated using steady-state chlorophyll fluorescence imaging and destructive biomass measurements. This combined evidence confirmed that Pst-induced chloroplast functional impairment strictly precedes macroscopic tissue structural collapse. This process effectively suppressed background noise while preserving critical infection-related signals. Subsequently, three classifiers-support vector machine (SVM), random forest (RF), and eXtreme gradient boosting (XGBoost)-were evaluated using the extracted 2D-COS features. Asynchronous features generally produced superior classification performance, with XGBoost achieving the highest accuracy (86.79%) and area under the receiver operating characteristic curve (AUC) (94.12%). Compared to conventional methods like PCA and CARS, 2D-COS more effectively attenuated growth-related interference and accentuated early disease signatures. These results demonstrate that the integrated framework of "multiplicative scatter correction (MSC) + Asynchronous Correlation Features + XGBoost" offers substantial potential for accurate, non-destructive, and early diagnosis of wheat stripe rust.