Arctic SAR Sea Ice Image Classification Based on Ice Charts ABSTRACT Sea ice is an important part of Arctic climate, which influences the global climate and environment significantly. Therefore, it is an urgent need to monitor the Arctic sea ice timely and accurately for the research of climatic change and the security of navigation. Synthetic Aperture Radar (SAR) has been the primary means to monitor the Arctic sea ice, since high-resolution SAR images can be obtained under all weather and day and night, meanwhile, SAR is also an important tool to acquire the Arctic sea ice distribution info rmation on a large scale. SAR sea ice image classification methods based on Markov Random Field (MRF) transform the classification task to the maximum a posteriori estimation problem, for which it is critical to choose the classification features with high robustness and model the classification features using MRF accurately. On the basis of analyzing deeply the expert knowledge provided by ice charts and the intrinsic property of sea ice, in the framework of MRF, two SAR sea ice image classification methods based on MRF are proposed using the prior knowledge at different levels in the dissertation, meanwhile the expert knowledge and the idea of between-class urrence relationship are incorporated. The main work of this dissertation is as follows: A multi-level SAR sea ice image classification b y incorporating egg-code-based expert knowledge is proposed, which makes use of the expert knowledge including the ice types, class numbers and partial concentration of sea ice existing in the specified egg code. According to the difference of partial concentration, the egg codes contained in the full SAR sea ice image are processed hierarchically. When region-level MRF segmentation is finished, the sea ice type labeling is realized bination with intensity feature. An arctic SAR sea ice image classification method based on between-class urrence features is proposed. Start from the i
