In order to explore the potential of estimating the fraction of absorbed photosynthetically active radiation (FPAR) of crops from unmanned aerial vehicle (UAV) multispectral remote sensing images, the vegetation index, texture index and leaf area index (LAI) were extracted from UAV multispectral images, which were used as model input parameters. On the basis of analyzing the correlation between different parameters and FPAR, the vegetation index and texture index were optimized. The FPAR of maize was estimated by unary linear regression (UL), multivariate stepwise regression model (MSR), ridge regression model (RR) and BP neural network model (BPNN). The results showed that the vegetation indexes, texture indices and LAI had a strong correlation relationship with FPAR, and the absolute value of correlation coefficient of vegetation index was the largest. Among different types of FPAR estimation models, BPNN model had the best estimation effect. The determination coefficient (R2) and root mean square error (RMSE) of FPAR estimation model were 0.857 and 0.173, respectively. The R2 and RMSE of validation model were 0.868 and 0.186, respectively. The relative error (RE) between model estimation value and field measured value was 8.71%. In different combinations of model parameters, the FPAR model fused with vegetation index, texture index and LAI had the best estimation and verification effect, which indicated that the fusion of multi feature parameters can effectively improve the estimation effect of FPAR. These results provided a scientific basis for precision estimation of maize FPAR and production potential based on UAV multispectral remote sensing data.