Abstract: Soil salinization in arid and semi-arid areas has a substantial negative influence on soil resources, eco-environment, and agricultural production. Quantitative inversion and monitoring of salinized soil can effectively protect land ecological security. Based on a spectral transformation method, indices of salinity characteristic bands and characteristic spectrum are derived, and salinity inversion models, including ridge regression (RR) and partial least square regression (PLSR) are constructed from measured hyperspectral and Sentinel-2B images. Characteristic spectrum indices are used as sensitive indices to match the Sentinel-2B images, the post-match salinity inversion models are constructed, and finally, soil salinity in Yinchuan Plain, Ningxia, China is quantitatively inverted and analyzed. The results show that the characteristic spectral indices of S₃ (Salinity index, S₃), Int₁ (Intensity index 1, Int₁), and Int₂ (Intensity index 2, Int₂) can realize the scale transition from measured hyperspectral cell to multispectral image pixel, which efficiently improves the accuracy of the salinity inversion model of multispectral image. The PLSR model after spectral matching can perform the best accuracy (R²=0.721, RMSE=4.856 g·kg⁻¹) of soil salinity inversion, with a 0.309 increment of R² and a 2.085 g·kg⁻¹ decrement of RMSE compared with the single Sentinel-2B image modeling. The result of salinity inversion is consistent with that from the field samplings, which demonstrates that characteristic spectrum indices contribute to spectral matching and fusion at different remote sensing image scales, and realize the quantitative monitoring of salinization from surface points to spatial dimensions, so the study could provide a theoretical and practical reference for soil salinity monitoring.