Abstract: The Shaanxi section of the Yellow River basin, located on the ecologically vulnerable Loess Plateau, acts as a significant agricultural production region in China. Accurate information extraction of cultivated land is of great significance for the evaluation of grain production capacity, the protection of cultivated land resources, and ecosystem management in the region. Based on the Google Earth Engine (GEE) platform and the multispectral remote sensing images from Sentinel-1 and Sentinel-2, this study created nine feature combinations, including spectral bands, remote sensing indices, texture features, radar features, and terrain features. Information on cultivated land was extracted using a random forest (RF) algorithm. Furthermore, 16 features with high contribution rates were selected through recursive feature elimination. Finally, random seeds were introduced to verify the system's robustness. The results indicate that the collaborative classification model based on multi-source features significantly outperformed the models using a single data source. When applied to information extraction of cultivated land in 2024, this model achieved an overall accuracy of 90.16% (Combination 9) and a Kappa coefficient of 0.802 0, representing an increase of 7.27% in classification accuracy compared to the model using spectral features only (Combination 1). Notably, terrain features effectively enhanced the identification results of cultivated land in areas characterized by great topographic fluctuations. The number of features was reduced from 26 to 16 through RF recursive feature elimination, significantly simplifying the computation while maintaining the classification accuracy. The optimized model achieved an overall accuracy of 92.00%, with a Kappa coefficient of 0.839 9. The morphological closing operation effectively eliminated the fragmented patches, significantly enhancing the continuity of patch boundaries. The method used in this study can provide a high-accuracy and generalizable technical framework for the dynamic monitoring of cultivated land in areas with complex terrain. Moreover, this method can provide support for decision-making in ecological protection and sustainable agricultural development in the Yellow River basin.