A Method for Classifying Land and Ocean Area by Removing Sentinel-1 Speckle Noise

Abstract

In Korea, satellite image-based land cover maps are limited because they are based on time-consuming pixel value-based classification techniques. The main categories of land cover classification are water and land; therefore, synthetic-aperture radar (SAR) images with high water reflectivity may be used to improve land cover map classification accuracy. In this study, C-band SAR images obtained by the Sentinel-1 satellite are used, which include various noises including speckle noise. To remove speckle noise, this paper applied Lee, Gamma, and Frost filters, and found that the Lee filter offered the best performance. By combining a stacking technique and the Lee filter, this paper successfully classified the target water system using image dichotomy and histogram analyses of the region of interest (ROI). The resulting land cover map showed 90 % accuracy compared to the pixel-based map, and comparison with the optical image showed that water coverage was effectively classified. This classification of forest reservoirs, which are difficult to distinguish in optical images, was rated as excellent. Thus, this speckle noise-removal technique will facilitate the improvement of land cover classification accuracy, particularly for flood boundaries and shorelines.