In order to meet the requirements of modern agriculture for dynamic, continuous, and rapid monitoring of crop growth, soybean was used as the research object based on UAV multispectral remote sensing technology in northwest China, and five vegetation indices were selected with the best correlation to soybean leaf area index (LAI), above-ground biomass and yield, and support vector machine (SVM), random forest (RF) and back propagation neural network (BPNN) were used to construct models for estimating soybean LAI, above-ground biomass and yield, respectively. RF and BPNN were used to construct and validate the models for estimating soybean LAI, aboveground biomass and yield, respectively. The results showed that the accuracy of soybean LAI and aboveground biomass prediction models constructed based on the RF model was significantly higher than that of SVM and BP models, with R2 of 0.801, RMSE of 0.675m2/m2, and MRE of 18.684% for the validation set of LAI estimation model; R2 of 0.745, RMSE of 1548.140kg/hm2, and MRE of 18.770. In the estimation model construction of yield, the soybean yield prediction model constructed based on RF model in soybean flowering period (R4) had the highest accuracy with R2 of 0.818, RMSE of 287.539kg/hm2 and MRE of 7.128 in the validation set. The research results can provide a theoretical basis for the application of UAV multispectral remote sensing in crop monitoring and provide a rapid estimation of crop yield application reference.