Low frequency microwave measurements have been used to gain insight into what happens deep inside ice sheets for some time now. In this paper, we used a deep neural network to classify each pixel within SMAP radiometer footprints over the Antarctica ice sheet as melt or no-melt. NASA's SMAP mission offers a valuable additional set of observations. The SMAP L-band (1.4 GHz) radiometer retrievals also cover virtually the entire Antarctica ice sheet twice a day. Consistent morning and evening sampling are provided by 6 AM/PM equator-crossings of the satellite ascending and descending polar orbits. The spatial resolution of the instrument is about 40 km. The cross-entropy loss function is used in our network. To make training and test sets, we used air temperature records from available weather stations to distinguish melt and no-melt ice sheet conditions. Our results show that the ice sheet experienced extensive surface melting during the 2015-2016 melt season, and also intensive melting in 2019-2020, particularity on the West Antarctic Ice Sheet.