A Novel Region-Merging Approach for Coastline Extraction From Sentinel-1A IW Mode SAR Imagery

Coastline extraction with high accuracy from wide-swath synthetic aperture radar (SAR) imagery is still a challenging problem due to speckle, sea state condition, and land type. This letter presents a novel approach for coastline extraction from SAR images with complex scenarios and large geograp hical coverage, using the combination of modified K-means method and adaptive object-based region-merging mechanism (MKAORM). First, a modified K-means method is used to produce initial oversegmentation for the following region-merging stage. Second, an adaptive and coarse-fine object-based region-merging scheme using subregion classification is exploited to extend the automatically selected “sea” seed and “land” seed, respectively, to extract the final coastline. MKAORM can reduce the high computation cost of coastline extraction from wide-swath SAR imagery while keeping high accuracy. More than 93% of the detected coastlines lie within 2-pixel distance in comparison with the manually traced coastlines in experiments taken on Sentinel-1A (S1A) IW (Interferometric Wide-swath) mode SAR imagery.