Abstract: n the underground mining process using the natural collapse method, due to the complex geological conditions such as high altitude and heavy rainfall, disasters such as mudslides or sudden mud surges often occur in the underground operation area, which poses a great threat to mining safety. To achieve rapid identification and early warning of the water content risk in the mining area, a soil moisture inversion method combining ground-penetrating radar and support vector regression (SVR) model is proposed. This method is a non-contact detection method. By introducing the measured radar data to extract echo signal features, a SVR prediction model is constructed and trained, thereby modeling the deep soil moisture. The results show that the model fitting accuracy reaches 0.9659, the test set error is less than 1%, and the relative error on the five randomly selected measurement points is less than 15%. Compared with the traditional neural network model, this model has higher accuracy and stronger robustness. This method can effectively meet the requirements of high-precision, real-time and non-contact moisture monitoring in complex mining areas, providing a feasible approach for improving the disaster identification ability and engineering safety level of underground mines.