Coastline Detection in Polarimetric SAR Images Based on GMM-HMM Segmentation Using Volume Scattering Component

It is difficult to detect coastline accurately in coastal zones containing extensive secondary echo sidelobe interference and complex coastal land covers in SAR images. A coastline detection method in polarimetric SAR images based on Gaussian Mixture Model-Hidden Markov Model (GMM-HMM) using volume scattering power is proposed in this article. To eliminate the interference caused by the secondary echo sidelobe of strong scattering buildings, the volume scattering power of a three-component decomposition is extracted for the sea-land segmentation. To describe the statistic distribution of the volume scattering power in heterogeneous coastal regions, the Gaussian mixture model is used in the Markov random field segmentation model. To obtain an efficient initial segmentation, a threshold segmentation method based on likelihood ratio in a combination of automatic sampling region selection is proposed. The distribution parameters of GMM-HMM are supervised estimated from the initial segmentation result. Quad-polarization data acquired by RADARSAT-2 in Dalian of China is used for the evaluation of the proposed method. Experiment results show that the average offset and the standard deviation are far less than those of the HMM coastline detection methods using scattering coherency matrix.