Harbor Detection in Polarimetric SAR Images Based on Context Features

The difficulties of harbor detection in SAR images lie in that the geometric features are blur and lack of commonality due to the diverse harbor morphology. A harbor detection method in polarimetric SAR images based on context features is proposed in this article. Firstly, extracting the polarimetric parameters of the three-component power components as input, the image is segmented to three regions including water, strong scattering built-up and the other regions using the Markov random field method based on Gaussian distribution. Then, based on the context feature of the harbor located on the coastline, the narrow-band region of interest (ROI) along the coastline is extracted from the water-land segmentation result for the harbor detection. Finally, based on the context feature that the man-made building in the harbor is a large area of strong scattering, harbors are discriminated from the ROI by determining the strong scattering region. Two quad-polarization data acquired by RADARSAT-2 in parts of Singapore and Zhanjiang, China demonstrate the effectiveness of the proposed method. The detection rate and false alarm rate of the proposed method are far better than a harbor detection method based on spatial geometric features.