UAV- and Machine Learning-Based Retrieval of Wheat SPAD Values at the Overwintering Stage for Variety Screening

In recent years, the delay in sowing has become a major obstacle to high wheat yield Jiangsu Province, one of producing areas China; hence, it is necessary screen varieties are resilient for late sowing. This study aimed provide an effective, fast, and non-destructive monitoring method soil plant analysis development (SPAD) values, which can represent leaf chlorophyll contents, late-sown winter variety screening. acquired multispectral images using unmanned aerial vehicle (UAV) at overwintering stage growth, further processed these extract reflectance five single spectral bands calculated 26 vegetation indices. Based on 31 variables, this combined three variable selection methods (i.e., recursive feature elimination (RFE), random forest (RF), Pearson correlation coefficient (r)) with four machine learning algorithms regression (RFR), linear kernel-based support vector (SVR), radial basis function (RBF) SVR, sigmoid SVR), resulted seven SVR models RFE-SVR_linear, RF-SVR_linear, RF-SVR_RBF, RF-SVR_sigmoid, r-SVR_linear, r-SVR_RBF, r-SVR_sigmoid) RFR RFE-RFR, RF-RFR, r-RFR). The performances 10 were evaluated compared each other according achieved determination (R2), residual prediction deviation (RPD), root mean square error (RMSE), relative RMSE (RRMSE) SPAD estimation. Of models, best was RF-SVR_sigmoid model, combination RF algorithm. It accuracy estimating values canopy (R2 = 0.754, RPD 2.017, 1.716 RRMSE 4.504%). newly developed UAV- learning-based model provided promising real time monitor contents stage, benefit