基于重轨星载SAR数据的多尺度快速变化检测

With the increasing of the acquirements and the development of the image resolution, the potential of space-borne SAR (Synthetic Aperture Radar) data is required to be fully studied. Driven by the repeat-pass space-borne SAR data and required to satisfy the demands for surveillance of pivotal regions on the globe, this project aims at improving the efficiency in use and the extent of automatic image interpretation. Through analyzing the characteristics of the space-borne SAR system, the high resolution SAR data acquired periodically are employed to fast and practically detect the changes of the regions of interest (ROIs), and it accordingly helps the experts to focus on the changed regions in the scenes which obviously enhance the efficiency of image interpretation. .There are mainly two parts in this research. One is the fast registration of the repeat-pass space-borne SAR data. Using DLS-EFM, the position of the satellite at any time can be estimated for the interference baseline calculation. Based on the indication of the baseline estimation, the elements of exterior orientation of the satellite and the imaging parameters of the sensor can be used for a fast and rough registration which is independent with the contents of the images. Then, a fine registration can be carried out in limited scopes which is beneficial for efficiency and precision from the rough registration. The other is the construction of the new change detector and the improvement of the thresholding method. The new change detector should be fast and make the differences between the changed and unchanged pixels significant, which can be used to generate an initial change detection map effectively. Then, with a sufficient consideration of the spatial relations between pixels and the polarimetric information, a change detection characteristic vector is constructed by introducing the multi-scale theory, and the fuzzy c-means algorithm (FCM) is adopted to generate the final change detection map by an iterative procedure, which should well and truly separates the changed and unchanged pixels with the original SAR data..

本项目基于重轨星载SAR数据驱动，面向全球重点区域监控的应用需求，旨在充分挖掘数量日益丰富、分辨率日益提高的星载SAR数据的侦监潜力，提高数据利用效率和图像自动解译程度。项目组拟利用周期性获得的高分辨星载SAR数据，实现重点监控区域快速、实用的变化检测，进而辅助判读员自动聚焦于感兴趣的变化区域，显著提高判读效率。研究内容主要包含两部分：第一，基于DLS-EFM对卫星空间位置进行估计，通过干涉基线估计的特点，利用卫星的外方位信息和传感器的成像参数，实现与图像内容无关的SAR图像快速概略配准，从而严格限制搜索区域，提高精配准的效率和精度。第二，在满足配准精度的基础上，利用基于图像灰度的快速变化检测算子生成初始检测结果作为迭代初值，然后引入多尺度分析算法，构建变化检测特征矢量，在充分考虑像素空间邻域信息和极化信息的基础上，采用模糊C均值算法进行聚类迭代，将变化像素和非变化像素准确、快速分离出来。

项目组主要围绕SAR图像的配准及变化检测方法展开研究，首先，对四种基于图像灰度的变化检测量进行了性能比较，然后利用其中最优的似然比变化检测算子提出一种SAR图像中人造目标自适应提取方法，并申请相关专利。第二，利用日本ALOS PALSAR数据星历表信息，基于椭圆轨道拟合的方法实现了卫星平台任意时刻位置亚米级的估计，并分析了卫星位置误差带来的地理编码误差的关系。第三，针对光学图像和SAR图像，提出了一种由粗到精的自动配准方法，基于图像直线特征，首先实现图像的概略配准，然后基于分块互信息，多次迭代，实现图像的精配准。第四，围绕多极化数据，提出一种基于四种散射模型的极化SAR图像变化检测方法，取得了良好的效果。最后，引入小波变换，提出一种多尺度融合的SAR图像变化检测方法，针对星载SAR图像对洪水灾害的情况进行了变化检测与评估。以上主要成果均已集成到SAR图像变化检测的软件原型系统中。