混合本征模型的多视SAR影像海冰密度检测

目的 SAR影像中像素光谱测度的空间相关性蕴含着海洋表面和海冰更加丰富的空间特性及其变化信息,因此合理建模这种相关性是高分辨率SAR影像海冰精准解译的关键。提出一种利用随机模型及空间统计学测度刻画海冰空间结构的方法。方法 本文首先,在空间统计学框架下,SAR影像被表示为多值Gamma模型和泊松线Mosaic模型线性加权构建的混合模型,其中多值Gamma模型用于描述海洋表面雷达信号背向散射变化的连续性,而泊松线Mosaic模型则用于表征不同类型海冰表面雷达信号背向散射变化的区域性。利用上述混合模型的一阶、二阶变异函数,建模蕴含在SAR影像中海冰空间结构的变化。结果 对RADARSAT-1影像海冰结构建模并反演其密度。实验区域真实海冰密度分别为20%,80%等,运用本文方法反演所得海冰密度与真实海冰密度误差正负不超过10%。结论 本文提出混合本征模型用以刻画SAR强度影像中海冰像素强度变化的空间关系,能够较好地反演Ungava湾海冰密度分布。为利用遥感影像检测空间机构提供一种全新的方法。

Intrinsic mixture model-based measurement of sea ice density from a multi-look SAR image

Objective The spatial structures revealed in remote sensing imagery are essential pieces of information that characterize the nature and scale of the spatial variation of sea ice processes. The freezing and melting of sea ice lead to changes in sea environment conditions, which in turn cause lockout, channel blocking, ship damage, and other issues. This study evaluates the potential capability of using the variogram of the intrinsic regionalization model to estimate sea ice density from synthetic-aperture radar (SAR) intensity images. Method A different geo-statistic metric is introduced, in which the spatial structures of sea ice are considered as a combination of two stochastic second-order stationary models. Under the stationarity assumption, a spatial structure model is proposed on the basis of second-order variograms to describe the sea ice density in multi-look SAR images. First, the multi-gamma model is used to characterize continuous variations that correspond to water or the background of sea ice. Second, a Poisson tessellation-based mosaic model is used, in which the image domain is randomly separated into non-overlapping cells. In each cell, a random value is independently assigned. The linear combination of these two stochastic models defines the mixture model to represent the spatial structures of sea ice presented in the SAR intensity imagery. Finally, the least squares method is used for the fitting method to estimate parameters. The image spatial structures are characterized by the variance weight and the variogram range related to each model. Result The proposed algorithm is applied to Radarsat-1 images acquired in different days to identify the change in sea ice. Experimental results show that the proposed method can accurately and stably estimate sea ice density. The real sea ice densities of the experimental areas are 20% and 80%. The errors between the real sea ice densities and the results from the proposed method are no more than plus or minus 10%. Conclusion In this study, the spatial structures of sea ice shown in a SAR intensity image are characterized by the intrinsic regionalization model. The algorithm is applied to a Radarsat-1 SAR intensity image to detect the sea ice change during two months in 2008. Results demonstrate that the algorithm is useful in detecting sea ice change in terms of intensity and size. However, these findings are limited by the number, types, and small size of the sea ice. Several aspects of the proposed algorithm can be improved in further studies. With the availability of the high-resolution polarization SAR image, the problem caused by high resolution and full polarization should be considered. For example, for the full-polarization SAR image, the spectral and spatial correlation of the same sea ice type becomes complicated. High resolution makes both the details and the noise in the homogeneous sea ice area prominent and enables the SAR images to show massive data characteristics. All of these issues result in unexpected difficulties in the design of high-resolution polarization SAR image-orientated algorithms for sea ice detection. In our future work, the suitability of the proposed algorithm to a high-resolution polarization SAR image will be improved through development of a spatial statistic model in spectral space.