Ice concentration estimation from dual-polarized SAR images using deep convolutional neural networks

High resolution ice concentration maps are of great interest for shipping and operations in the Arctic. A convolutional neural network (CNN) has been used to estimate ice concentration from C-band dual-polarized RADARSAT-2 satellite images in the melt season. SAR images are used as input and the ice concentration is the direct output from the CNN. With no feature extraction or segmentation-based post processing, the errors of the generated ice concentration maps compared to image analyses are less than 10% on average. The CNN is demonstrated to produce ice concentration maps with more details than image analyses. Reasonable ice concentration estimations are made in the melt season, and regions of low ice concentration ice are also well captured.