An improved method for SAR image coastline detection based on despeckling and SVM

Coastline detection in synthetic aperture radar (SAR) image is an important component for SAR image interpretation. However, because of speckling, SAR image coastline detection with high accuracy is far from resolved. In the paper, an improved method using multiscale wavelet-based despeckling and support vector machine (SVM) based classification is proposed for SAR image coastline detection. In the method, as a pre-process step, a multi-scale wavelet SAR image despeckling strategy is proposed to suppress the noise of SAR image. Base on despeckled SAR image, a novel method for water and non-water classification using circular-window Gray Level Co-occurrence Matrix (GLCM) and SVM is proposed. GLCM of circular-window is designed as texture feature of water and non-water areas for classification. Feature vector of eighteen dimensions is derived from GLCM and fed into a SVM-based classifier to get water region, the contour of water region is extracted as coastline. The experimental results using real SAR images demonstrate that the proposed approach has better performance compared with other ones.