Abstract: Objectives:As a sensitive indicator of global change such as ice sheet mass loss, sea level rise, climate change and so on, supraglacial lakes (SGLs) play an important role in the study of ice sheet stability under the background of global climate change. Methods:The semantic segmentation model U-Net in deep learning is used to extract SGLs from Landsat satellite imagery in different situations, and compared with many other methods. Results:It is found that U-Net supraglacial lakes extraction method possesses the optimal performance and efficiency. Meanwhile, we extract the SGLs near the Sermeq Avannarleq Glacier in Greenland Ice Sheet from 2000 to 2020, and the seasonal and interannual variations of their areas are analyzed. Conclusions The SGLs area starts to increase in mid-May, reaches the peak in the mid-late-July, and almost disappears in September. As for the long time-series change, the area of SGLs in this region shows an increasing trend from 2000 to 2020, indicating that the amount of ice sheet melting is gradually increasing.