Arctic Ocean Surface Type Classification Using SAR Images and Machine Learning Algorithms

This work is aimed at the investigation of the application of general machine learning (ML) algorithms, used typically for image recognition, to ocean surface type categorizing, using Synthetic Aperture Radar (SAR) images as input data. The study is focused on the Convolutional Neural Network (CNN) based image segmentation. Its main goal is building the fully automatic pipeline, calculating (predicting) ocean surface (particularly ice) type from a SAR image.It is shown that the developed system has classification accuracy of up to 90% without any preliminary feature selection or region-based limitations. The advantages and drawbacks of the suggested approach are discussed.Sea ice - sea water discrimination and sea ice concentration estimation from Sentinel-1 SAR images are considered in more details due to their general availability and extensive coverage of the Arctic region.