We input remote sensing images into the network, and after a lot of convolutional layers, the network will output a segmentated images with white part showing the glacier and black part showing the ocean. With that segmented images, we can easily extract the glacier terminus.
After training with huge amont of training data, our deep learning network can handle various situations and produce very decent glacier termini.
The automation greatly improves the temporal resolution and spatial coverage of the terminus data. With these fine-resolution time series of terminus variation, we now can conduct exciting research about ice-ocean interaction in great detail!