SAR Sea Ice Image Segmentation Based On Ice Chart

Sea ice has significant scientific value in global climate research as a sensitive indicator of global climate variability, and also has great strategic value in global economy pattern transformation as the navigation capability of the "golden waterway" has been gradually enhancing, which contain the Northwest Passage and the Northern Route. Synthetic aperture radar (SAR) images are now playing a key role in sea ice studying. As SAR sea ice images were spatio-temporal non-stationary (different times or in different locations, statistical characteristics may be different), automatic interpretation entirely by computer is poor in the robustness, and need to be complemented by some expert knowledge. Ice charts contain rich expert knowledge, such as the type, concentration and floe size of ice, and has a wealth of historical archive for free use, but only a limited number of studies started exploiting the information capture in these data. By combining the two, this thesis will introduce the expert knowledge in ice charts into the process of SAR sea ice segmentation, and propose a new approach for SAR sea ice image segmentation.The main work of this thesis is as follows:(1) By employing georeferenced SAR remote sensing data of the Canadian satellite RADARSAT-1, this thesis established the conversion between the original radar radiometric intensity and the backscatter coefficient, and obtained the SAR images in common sense.(2) By employing the free ice charts of Canadian Ice Service, this thesis established the transformation between the geographical coordinates in ice charts and the grid coordinates in SAR images, and achieved the unification of both coordinate systems.(3) Taking MRF-based image segmentation method as the basis, and the expert knowledge extracted form ice charts as guides, the thesis proposed a new approach which combines the expert knowledge and automatic segmentation, and constructed a SAR sea ice segmentation system based on ice charts.(4) In light of the lack of basic research platform and the difficulty of follow-up study, the thesis accomplished a basic object-oriented library and a user interface based software.The experimental results have showed the higher performance in both segmentation result and speed,and demonstrated the efficiency of the proposed method.