In order to clarify the characteristic spectrum of main salt ions in arid areas, a prediction model for high-precision and stable saline soils was established.Taking Fukang City of Xinjiang as the study area, collected 55 soil samples and field measured spectral data based on VIS-NIR, using multiple linear regression(MLR), support vector machine(SVM) and random forest(RF) method three inversion model of soil salinity and main ion content were established, and the model was tested. The results showed that: 1) At 0.01 significant level, soil salinity had a significant correlation with Na⁺, Cl^- and Ca²⁺, and the correlation coefficients were 0.978, 0.814 and 0.645, respectively; 2) Comprehensive spectrum response and correlation analysis determined the dominant ion bands of soil salt at 459, 537, 1381, and 1 386 nm, and the significant characteristic bands at 459 and 537 nm; 3) The three model fitting effects from high to low were RF>MLR>SVM in order, and using the model established by RF, the salt main ions (Na⁺,Cl^-,Ca²⁺) had the highest R², the smallest RMSE, and the largest RPD, which were 2.11, 2.03, and 1.80, respectively, and were the optimal prediction models. By selecting the dominant characteristic bands of major ions in the soil, RF method was used to construct the estimation model in this area, which can effectively extract the main ion information of soil salinity in the arid area.