Water resources have become a key factor for restricting the social, economic and agricultural development of arid area in Northwest China. In recent years, agriculture in arid oasis has developed rapidly, and human activities have seriously affected balance on the regional soil moisture, resulting in a large area of salinization. Therefore, the monitoring of soil moisture is of great practical significance to the development of oasis agriculture and economy. Taking the oasis of Weiku in Xinjiang as the study area, totally 41 soil moisture samples and hyperspectral data of the oasis vegetation in arid area were collected, and the vegetation index was taken as bridge. Multiple regression (MLSR), partial least squares (PLS) regression and support vector machine regression (SVR) were used to establish the inversion model of soil water content in oasis, respectively, the regression models were tested respectively. The experimental results showed that the accuracy of different models was different. Through the optimization of parameters and extraction of optimal test set, the fitting effect from good to bad was improved SVR model, PLS model and MLSR model, which were based on the vegetation The improved SVR model had a good fitting effect, R2 was 0.8916, RMSE was only 2.004, the analysis accuracy in the oasis of arid area reached the practical prediction accuracy. The R2 values of MLSR model and PLS model were 0.6300 and 0.6549, and RMSE were 3.001 and 2.749, respectively. The results showed that it was an effective method to improve the monitoring accuracy of shallow soil water content in oasis, and it can also provide more data for monitoring soil moisture in arid area.