Sea-land Segmentation in Polarimetric SAR Images

With the development of SAR technology, quad-pol SAR has been utilized for multiple scenarios for its rich polarization information. To verify the potential of quad-pol SAR in the sea-land segmentation assignment, we adopt superpixel, random forest, and UNet neural networks from the perspective of methods. Based on the dataset produced from Gaofen-3 quad-pol SAR images, experimental results show that multi-polarization information can improve the sea-land segmentation accuracy under the same algorithm. Besides, the UNet method has a better performance than superpixel and random forest on both accuracy and time consumption.