Soil moisture is an important factor affecting agricultural production and plays a vital role in crop growth and final yield. Rapid and efficient estimation of soil water content has become a hot issue in agricultural and forestry water resources monitoring. It has been widely recognized and applied to calculate vegetation index and build soil water content inversion model by using the characteristic bands of hyperspectral reflectance. Because of the problem that the inversion of soil water content is greatly affected by vegetation coverage, we propose to use multi vegetation index combination to weaken the influence of vegetation coverage on the inversion of soil water content. Thirty groups of citrus trees were selected as samples in the Cangwubang test base of Yichang City. The soil was collected at the drip line of the fruit tree, and the soil mass moisture content was determined by the drying method. Four times of sampling, a total of 120 groups of soil moisture content. We use the ASD Field Spectral FR spectrometer (wavelength range: 325~1 075 nm) and the Dajiang Genie 4 multispectral UAV to obtain the spectral reflectance in the blue, green, red, red edge, near-infrared and short wave infrared bands of 120 groups of test areas. We pretreat the spectral data with the moving average method for noise reduction, compare and analyze 9 vegetation indices with gray correlation method, and screen out 4 vegetation indices that are highly significantly related to soil water content (p<0.01). The correlation between each index and soil water content from high to low is the bare soil index (BSI), normalized blue-green differential vegetation index (NGBDI), green normalized index (GNDVI) and normalized differential vegetation index (NDVI). The correlation between BSI and soil water content is the highest, and the correlation coefficient is -0.687. We use the linear stepwise regression method and nonlinear BP neural network method to build a soil water content inversion model based on multi vegetation index and take the determination coefficient (R2), root mean square error (RMSE) and relative error (ARE) as the evaluation indexes of the inversion accuracy of the model. The results show that the R2 between the inversion value of soil water content and the measured value of the stepwise regression model and BP neural network model are 0.816 and 0.889 respectively, the RMSE is 2.54% and 1.53% respectively, and the ARE is 21.13% and 8.88% respectively. It shows that the nonlinear BP neural network algorithm based on multi vegetation index combination has higher accuracy in soil moisture inversion based on vegetation index modeling, and can overcome the influence of different vegetation coverage on the accuracy of soil moisture inversion to a certain extent. As an effective alternative method to measure soil moisture directly, it provides theoretical support for quantitative decision-making and scientific agricultural irrigation management.