Rivers play an important part of the earth's water cycle， and it is significant to extract mountain river information accurately for water resource evaluation and ecological restoration. According to Sentinel-2 images from 2019 to 2021， and the multi-spectral index method was used to distinguish rivers， lakes and glaciers automatically by combining with the Random Forest （RF）， and MERIT DEM was used as the terrain condition to extract the multi-temporal and high-resolution river automatically. Combined with the second glacier inventory dataset of China， the catchment of different recharge types were divided， the percentage change in river area and width was further calculated to describe the seasonal changing rate. The results showed that the average area of the river in the wet season and the dry season has reached 7 161.64 ± 22.73 km² and 4 066.02 ± 35.19 km² during 2019~2021， respectively， and both the Kappa coefficient and the overall accuracy are larger than 0.8. The average seasonal changing rates of river areas of glacier-fed types and non-glacier-fed were 0.34 and 0.23， respectively. The seasonal variation of the average width of glacial-fed rivers was mostly larger than that of non-glacial-fed rivers. The results showed that the changes in width and area of glacier-fed rivers are much larger than that of non-glacier-fed rivers， suggested that glacial meltwater has an important impact on the changes of river runoff.