Impervious surface is an important indicator for analyzing urban expansion, measuring the degree of urbanization, and characterizing the urban ecological environment. Accurately extracting impervious surface data is of great significance to regional economic development, disaster prediction, ecological restoration, and environmental assessment. In this paper, using multi-spectral, synthetic aperture radar (SAR) and surface temperature retrieval (LST) data, from the four perspectives of spectral features, time series features, SAR texture features, and coherence features, an index feature that highlights impervious surface information is constructed. The model generates a 10m impervious surface product in the urban area of Shaoguan, and verifies the accuracy of the real sample based on the 0.8m GF-2 data in the same period. The results indicate that the overall accuracy of extracting impermeable surfaces in the study area is 94%, with a Kappa coefficient of 0.92. The multi feature fusion random forest model combining optical data and SAR data has high extraction accuracy and applicability in the mountainous areas of southwestern China, which improves the misclassification of bare land and other land types as impermeable surfaces.