Abstract: Abstract: The prevention and control of saline soil is of great significance to the sustainable development of agriculture in Xinjiang, and comprehensive knowledge of the Salt content of saline soil is a prerequisite for the prevention and control of salinization, so it is necessary to find an accurate and rapid method to estimate salt content information of saline soil. Ground Penetrating Radar (GPR) is widely used in geophysical and environmental assessment as a non-destructive detection technology. In recent years, this technology focuses on the detection of water content. However, the detection of soil salt is relatively weak due to the large dielectric loss in saline-alkali soil, the easily-absorbed electromagnetic wave energy the complex way to process radar signal, and even the different frequency combinations are needed to test, all of which above bring great troubles to the detection of salt. However, the rapid development of dielectric models in salinized soil makes it possible to use this technique to retrieve salt content. In order to better understand the soil salt content in drip irrigation cotton field, a typical drip irrigation cotton field was selected as the research object to be detected with a Canada's Ekko PRO series GPR multi-frequency antenna (250 MHz and 1 000 MHz). Then, the soil dielectric constant was obtained through radar amplitude envelope Average method (AEA). Soil salt content was estimated by adopting the Dobson saline soil dielectric model (Dobson dielectric model of saline soil, Dobson) as the theoretical tools. Meanwhile, the dielectric constant, soil bulk density, water content, and soil clay content were applied as the input variables of the model; Multiple Linear Regression (MLR) method was used to establish soil salinity inversion model; BP neural network (BP) was employed for simulation and prediction. Finally, the accuracy of salt content inversion by MLR model, BP model and Dobson model was evaluated based on the measured salt content. The results showed that: (l) The effective depth of GPR 250 MHz and 1 000 MHz frequency antenna AEA method was 0-30 cm. (2) The dielectric constant obtained by 1 000 MHz frequency antenna AEA had a good polynomial relationship with the measured water content, and the fitting effect and accuracy of the measured water content and the inverse water content were good (R2, RMSE, MER were 0.96, 1.61%, and 7.25%, respectively). (3) Among the three salt inversion models, the inversion accuracy of Dobson salinized soil dielectric model was significantly higher than that of the other two methods, with R2 and RMSE of 0.91 and 0.313 g/kg respectively. Therefore, it is suitable and reliable to use GPR multi-frequency antenna AEA method to estimate soil salt content in cotton fields under drip irrigation. This method provides a new way to retrieve soil salt content and enriches the methods and means of detecting soil salt content.