The accurate extraction of impervious surface is of great significance for regional population density estimation， environmental assessment， disaster prediction， hydrological model construction， urban heat island effect research and climate change analysis. Traditional large scale impervious surface extraction methods are mainly limited by the quality of remote sensing data and the selection of extraction features， and the spatial resolution of extracted impervious surface is low， which is difficult to meet the refined requirements of impervious surface at the present stage. In this paper， based on Sentinel-1 SAR and Sentinel-2 MSI remote sensing data， multiple extraction features of impervious surface were selected from three dimensions， including spectrum， texture and time sequence， to build an impervious surface extraction model based on random forest. In addition， GEE platform was used to carry out extraction experiment of 10m impervious surface in Yangtze River Delta region in 2020. The results showed that in different types of experimental areas， compared with spectral features， spectral features and time series features， the overall accuracy and Kappa coefficient of the proposed method were increased by 5%，9% and 2%，6%， respectively， and all cities with different impervious surface coverage levels had good extraction effects. The overall accuracy and Kappa coefficient of impervious surface extraction at the global scale in the Yangtze River Delta region were 93.75% and 0.88， respectively. The impervious surface area was 6 1591.38 km²， accounting for about 17% of the total area. The impervious surface extraction method proposed in this paper for 10m resolution remote sensing images is suitable for different types of areas such as mountainous areas， rural areas， urban areas and urban fringe areas. The method is simple and easy to operate， has high precision， and is suitable for cloud platform large-area computing.