The accuracy of the data captured by UAV multispectral remote sensing for winter wheat yield prediction is still not high, and in order to guide the accurate prediction of winter wheat yield at the field scale, a high-precision winter wheat yield estimation model needs to be constructed. The corrected near-ground hyperspectral data (acquired by Field-Spec 3 analytical spectral devices, ASD) was used to verify the low-altitude UAV multispectral remote sensing data (acquired by DJI Phantom 4 multispectral camera, P4M), and the vegetation index calculated by the UAV multispectral image was combined with empirical statistical methods, and unvariate regression and multiple linear regression were used to estimate yields based on a single vegetation index and the combination of multi-vegetation index at the panicle stage, flowering stage and filling stage, respectively. Among them, the combination of multi-vegetation index included the normalized difference vegetation index (NDVI), the optimized soil adjusted vegetation index (OSAVI), the green normalized difference vegetation index (GNDVI), the leaf chlorophyll index (LCI) and the normalized difference red edge index (NDRE). The results showed that the winter wheat yield estimation model based on a single vegetation index had the highest accuracy, while the multiple linear regression model based on five vegetation indices had better fitting effect than the single vegetation index model in the three growth periods. Univariate or multiple regression models fit best during the spike extraction period. The coefficients of determination (R2), root mean square error (RMSE) of the modeling set of winter wheat based on the GNDVI index of the univariate quadratic regression yield estimation model were 0.69 and 428.91kg/hm2, respectively, and the R2, RMSE and relative root mean square error (RRMSE) of the validation set were 0.76, 418.14kg/hm2 and 11.56%, respectively. The R2 , RMSE and RRMSE of modeling set of the multiple linear regression yield estimation model based on the combination of five vegetation indices were 0.80, 340.14kg/hm2, and the R2, RMSE and RRMSE of the validation set were 0.69, 466.75kg/hm2 and 12.90%, respectively. In summary, the data captured by the P4M had broad application prospects in estimating winter wheat yield. The optimal model for winter wheat yield estimation was a multiple linear regression model based on the combination of multiple vegetation indices at the ear pumping stage.