Hybrid strategy for precise sea-land segmentation in GF-3 SAR images

Gaofen-3(GF-3) was the first multi-polarization radar satellite launched by China which is dedicated to marine monitoring and ocean remote sensing. Sea-land segmentation is a vital step of image preprocessing. In this paper, we propose a novel hybrid strategy for precise sea-land segmentation in GF-3 SAR images, which can improve the robustness of segmentation performance with respect to speckle noise, waves, imaging, and etc. Three different strategies are adopted in different polarized channels:1) logarithmic Gaussian mixture model(LGMM) is applied to distinguish all-land and all-sea images from mixture images by determining the ratio of sea-to-land, and it conducts fine clustering of sea-land segmentation according to the parameters of LGMM; 2) OTSU method, which maximizes inter-class variance in logarithmic domain and complex domain respectively, and generates the sea-land masks based on gray information; 3) logarithmic cumulants are adopted to analyze texture information, and judges the general distribution of mixture images by the connection between logarithmic second-order moment and logarithmic third-order moment, so as to verify and predict the sea-land masks generated by majority voting mechanism, and enhance the accuracy of sea-land segmentation. Experiments were carried out on a large number of GF-3 SAR images, and the results are exceptional. The proposed sea-land segmentation algorithm has potential practical application.