Modeling Sea Ice Texture Structure From SAR Image Based On Spatial Statistics

The spatial information can be well characterized by remote sensing imagery.With the development of remote sensing,SAR(Synthetic Aperture Radar)imagery is widely used in spatial data analysis.But because of the complexity of its imaging process,it is difficult to interpret the original data.In this paper,on the basics of stochastic geometry and spatial statistics,we use geostatistic metrics and stochastic models to characterize the sea ice spatial structures.First we build two stochastic models according to the properties of the sea ice data.One is a multi-Gamma model,which characterizes continuous variations corresponding to water or the background of sea ice.The other is a Poisson line mosaic model,which characterizes the regional variations of different types of sea ice.Then based on the two models,we build the mixture model and define its geostatistic metrics——theoretical first-and second-order variograms to characterize the variation of sea ice spatial structures.At last,to estimate the parameters of the mixture model,experimental first-and second-order variograms are calculated from the SAR intensity imagery,and then fit them with the theoretical variograms for the purpose of estimating the mixture model parameters:The experiment has three sections.First we verify the feasibility of the mixture model of discriminating different spatial structures.Then simulated images are generated from the two stochastic models,they are used to verify the accuracy of this method.In the end,the proposed method is applied to Radarsat-1 images from April to June to identify the change of sea ice.The results of the experiments show that the proposed approach can estimate the sea ice density accurately and stably.