Remote Sensing Monitoring of Winter Wheat Stripe Rust Based on mRMR-XGBoost Algorithm

For the problem of multi-dimensional feature redundancy in remote sensing detection wheat stripe rust using reflectance spectrum and solar-induced chlorophyll fluorescence (SIF), a selection disease index (DI) monitoring model combining mRMR XGBoost algorithm was proposed this study. Firstly, characteristic wavelengths selected by successive projections (SPA) were combined with vegetation indices, trilateral parameters, canopy SIF parameters to constitute initial set. Then, max-relevance min-redundancy (mRMR) correlation coefficient (CC) analysis used reduce dimensionality set, respectively. Features CC input as independent variables into extreme gradient boosting regression (XGBoost) tree (GBRT) monitor severity rust. The experimental results show that, compared analysis, accuracy features GBRT models increased 12% 17% on average, Meanwhile, mRMR-XGBoost achieved best (R2 = 0.8894, RMSE 0.1135). R2 between measured DI predicted improved an average 5%, 12%, 22% mRMR-GBRT, CC-XGBoost, CC-GBRT models. These suggested that is more suitable for rust, has advantages than commonly selection. Field survey data validation also confirm excellent applicability scalability. could provide reference reduction crop based hyperspectral data.