Title:Coastline extraction from ALOS-2 satellite SAR images

Abstract:The continuous monitoring of a shore plays an essential role in designing strategies for shore protection against erosion. To avoid the effect of clouds and sunlight, satellite-based imagery with synthetic aperture radar is used to provide the required data. We show how such data can be processed using state-of-the-art methods, namely, by a deep-learning-based approach, to detect the coastline location. We split the process into data reading, data preprocessing, model training, inference, ensembling, and postprocessing, and describe the best techniques for each of the parts. Finally, we present our own solution that is able to precisely extract the coastline from an image even if it is not recognizable by a human. Our solution has been validated against the real GPS location of the coastline during Signate's competition, where it was runner-up among 109 teams across the whole world.