In order to make up for the defects of the one-time modeling analysis and improve the operation efficiency and accuracy of wheat stripe rust remote sensing detection model, based on the characteristics of model population analysis (MPA) algorithm and the advantages of spectral interval selection algorithm and spectral point selection algorithm, a feature variable selection algorithm was proposed, combining correlation coefficient (CC) and MPA. Based on the selection of feature variables by CC algorithm for the full band spectrum, competitive adaptive reweighted sampling (CARS) and variable combination population analysis (VCPA) developed based on MPA were used to further optimize the feature variables sensitive to wheat stripe rust, and partial least squares regression (PLSR) algorithm was used to construct CC-CARS and CC-VCPA models for remote sensing monitoring of wheat stripe rust. The results showed that the accuracy of CC-CARS and CC-VCPA models constructed by combining the feature variables selected by CC-MPA algorithm was higher than that of CC, CARS and VCPA algorithm. In the three groups of validation set samples, CC-CARS model compared with CC model and CARS model, the R2V between predicted disease index (DI) and measured DI was increased by at least 6.78% and 6.66%, RMSEV was decreased by at least 15.31% and 10.98%, and RPD was increased by at least 18.08% and 12.34%, respectively. Compared the CC-VCPA model with CC model and VCPA model, the R2V between predicted DI and measured DI was increased by 9.58% and 0.73%, RMSEV was decreased by 20.78% and 3.86%, and RPD was increased by 26.22% and 4.02%, respectively. The spectral feature optimization algorithm based on CC-MPA was an effective feature selection method. In particular, the number of feature variables selected by CC-VCPA method was less and the model prediction effect was better. The research results had important reference value for spectral feature optimization and improving the accuracy of remote sensing monitoring of crop diseases.