Soil salinization, a major environmental and socioeconomic challenge worldwide, is jointly influenced by human activity and climatic conditions. Therefore, the aim of this study was to investigate the spatiotemporal dynamics of soil salinization in the semi-arid irrigated region of the upper and middle reaches of the Yellow River, specifically the Yinchuan Plain (YP), from 2005 to 2022. Using Landsat-5 and Sentinel-2 satellite data and extensive surface soil salinity measurements, soil salinity inversion models were established for 2005, 2009, 2017, and 2022 (Validation set: Landsat-5, R = 0.81, RMSE = 24.2 g/kg, MAE- 13.68 g/kg; Sentinel-2, R = 0.86, RMSE1.94 g/kg, MAE = 1.34 g/kg). The results showed that the low salinization level was dominant, and that the nonsalinized area increased from 1385.3 km2 in 2005 to 2743.6 km2 in 2022. Most of the changes occurred among non-salinized, low-salinized, and medium-salinized areas, with the proportion of the non-salinized area increasing from 22.31% to 42.81% and the total salt-affected area decreasing from 45.15% to 33.26%. Cropland was the main land cover type and the most affected by salinization, with low salinization being dominant. Notably, spatial variability in prediction uncertainty was observed (e.g., high uncertainty in the northwestern corner in 2005 and southern region in 2009), which is acknowledged to ensure robust interpretation of the findings. In this study, a long-term measured salinity dataset was integrated with machine learning inversion techniques to quantify salinization transitions across a 17-year period in the YP. The findings provide insights for the evaluation of soil salinization in irrigated semi-arid regions.