U-Net for Taiwan Shoreline Detection from SAR Images

Climate change and global warming lead to changes in the sea level shoreline, which pose a huge threat island regions. Therefore, it is important effectively detect shoreline changes. Taiwan typical island, located at junction of East China Sea South Pacific Northwest, deeply affected by In this research, was selected as study area. an efficient detection method proposed based on semantic segmentation U-Net model using Sentinel-1 synthetic aperture radar (SAR) data island. addition, batch normalization (BN) module added convolution layers architecture further improve generalization ability accelerate training process. A self-built dataset introduced train test its efficiency. The consists total 4029 SAR images covering all coastal areas Taiwan. samples were annotated morphological processing manual inspection. results then processed edge postprocessing extract shoreline. experimental showed that could achieve satisfactory performance compared with related methods provided Ministry Interior from 2016 2019 for different landforms Within 5-pixel difference between detected ground truth data, F1-Meaure exceeded 80%. potential validated sandbar southwestern coast Finally, entire has been described approach length close actual length.