Sobel 엣지 검출과 전자해도를 이용한 KOMPSAT-5 이미지의 개선 된 육상 마스킹 방법

Abstract

The southern part of Korean peninsula consists of many small islands and being used as routes of many passenger ships. Therefore, detection of lands, particularly detection of small islands is necessary in synthetic aperture radar (SAR) images in order to avoid misidentification of those small islands as ships due to their small size and similar reflectance properties. Yet now there are many developed land masking processes for SAR images which usually use the freely available static coastline information but those usually exclude small islands. Therefore, in this study we have used moderate resolution X-band HH polarization KOMPSAT-5 (K-5) images, and very high resolution coastline data from electronic navigation chart (ENC) in order to accurately detect the small islands. We used Sobel edge detection which was found to properly detect even the irregular shaped coastlines from binary K-5 images. The broken edges were connected together to complete the coastline which were produced from applying high threshold values to the whole K-5 scenes. The regions inside the produced coastlines were filled up to create a temporary land mask which also included objects other than lands. Those temporary masks were geocorrected, and separate land areas were marked differently using object labelling method in MATLAB. In order to deduct the objects other than lands, pixel DN values were multiplied with the ENC-based coastline map cges in order to avoid misidentification of those small islands as ships due to their small size and similar reflectance properties. Yet now there are many developed land masking processes for SAR images which usually use the freely available static coastline information but those usually exclude small islands. Therefore, in this study we have used moderate resolution X-band HH polarization KOMPSAT-5 (K-5) images, and very high resolution coastline data from electronic navigation chart (ENC) in order to accurately detect the small islands. We used Sobel edge detection which was found to properly detect even the irregular shaped coastlines from binary K-5 images. The broken edges were connected together to complete the coastline which were produced from applying high threshold values to the whole K-5 scenes. The regions inside the produced coastlines were filled up to create a temporary land mask which also included objects other than lands. Those temporary masks were geocorrected, and separate land areas were marked differently using object labelling method in MATLAB. In order to deduct the objects other than lands, pixel DN values were multiplied with the ENC-based coastline map c