Sea and ice discrimination and classification in the polar regions from satellite data gained
importance in remote sensing and geosciences, essentially because of the undergoing climate
change. Synthetic Aperture Radar (SAR) is one of the best instruments in remote sensing for
sea-ice discrimination at high spatial resolution (hundreds of m), because it provides images
day and night with whatever cloud coverage. Sentinel-1 is a European Space Agency (ESA)
satellite mission designed for Earth observation, specifically focused on monitoring changes
in the Earth’s surface. Launched in 2014, the Sentinel-1 satellite uses SAR in the C band,
which is why the Sentinel-1 data are used in this thesis. The area of interest used in this
thesis is a small part of Norway Island in the north of Europe. This area is cold because it is
close to the polar region, and most of the time there is ice in the area. The available ice masks
are generally at spatial resolution of km, thus not fully suitable to be used with SAR images.
This thesis studies the sea and ice discrimination in high resolution with good accuracy. The
target in this thesis is to employ unsupervised techniques due to the absence of training data
and the difficulty in evaluating such samples. In this thesis two different approaches are in-
troduced, the first based on texture feature extraction and the second based on deep learning
methods. The discrimination of the ice from sea is with the existence of wind, which make it
the challenging conditions. The texture feature extraction approach is based on three texture
analysis methods, which are GLCM, MLBP and Tamara, with over-segmentation to reduce
the noise effects and FCM for discrimination. The deep learning-based approach is based
on the cGAN network, which is trained with sea textures and is able to generate only sea,
and the post-processing of anomaly detection. At the end, the results show the proposed ap-
proaches have better quality than available ice masks an