Abstract: Due to the low prediction accuracy of existing SAR sea-land segmentation, the segmentation network model adopted is generally large and difficult to deploy on board. A binarized conditional random field segmentation network (BiCSNet) for polarimetric SAR is proposed. The lightweight model is mainly realized by the binarization convolution module designed for the binary task of sea-land segmentation. In order to improve the segmentation accuracy of the lightweight network, the convolution conditional random field is incorporated into BiCSNet to realize the end-to-end network prediction. Based on the data set constructed from quad-polarization SAR images in coastal areas of China, the proposed network has good performance in both accuracy and lightweight.