基于相干图学习的干涉SAR图像分割

发布时间：2018-05-29 08:17

本文选题：干涉SAR + 相干图　；参考：《西安电子科技大学》2014年硕士论文

【摘要】：干涉SAR(Interferometric Synthetic Aperture Radar,InSAR)作为合成孔径雷达的延伸和发展,主要关注于获取数字高程图(Digital Elevation Model,DEM)。本文在介绍干涉SAR技术的基础上,着重于干涉SAR图像的分割。相干图作为干涉SAR技术的中间步骤,有着重要的物理意义。它不仅作为相位图的评价标准,而且对于地物具有良好的可分性能。本文在对相干图的特性进行分析的基础上,利用其特性进行地物划分。然后利用相干图的良好的可分性,将其迁移到目标SAR图像上,以获取更好的SAR图像的分割效果。本文的主要工作如下:1.提出了一种基于相干特性与空间信息的干涉SAR的相干图分割方法。首先对相干图的同质区域进行统计分析,比较不同匀质区域的均值和方差参数,基于相干图的统计特性,利用最大后验概率的贝叶斯分类器进行相干图的初始划分,同时引入Markov随机场,利用相干图的邻域信息,从而有效的进行杂点去除。实验结果显示挖掘出的相干图特性和空间信息能够较好实现干涉SAR的相干图分割。2.提出一种相干图迁移聚类的SAR图像分割算法。利用Kmeans方法对目标SAR图像进行初始划分,根据得到的初始标签,用K-SVD算法为每一类数据学习一个字典。为了减少计算的复杂度和降低算法的随机性,根据每一类的聚类中心,选择距中心最近的一批样本作为训练数据来训练字典。在进行目标SAR图像分割时,将相干图中可分的信息通过稀疏表示的字典,迁移到目标SAR图像上,两个真实干涉SAR图像实验表明分割的效果得到明显提升。本工作得到了国家自然基金(Nos.61003198,61472306)和中央高校基本科研业务费专项资金(JDYB140508)的资助。
[Abstract]:As an extension and development of synthetic Aperture Radar (SAR), interferometric SAR(Interferometric Synthetic Aperture Radarar (Insar) is mainly concerned with obtaining digital elevation map (DWT). Based on the introduction of interference SAR technology, this paper focuses on the segmentation of interference SAR images. As the intermediate step of interference SAR technique, coherent graph has important physical significance. It not only serves as the evaluation standard of phase map, but also has good separability for ground objects. In this paper, based on the analysis of the characteristics of coherent maps, the features of ground objects are divided. Then the coherent image is migrated to the target SAR image by using the good separability of the coherent image to obtain a better segmentation effect of the SAR image. The main work of this paper is as follows: 1. A coherent image segmentation method based on coherent characteristics and spatial information for interference SAR is proposed. Firstly, the homogeneity region of coherent map is statistically analyzed, and the mean and variance parameters of different homogeneous regions are compared. Based on the statistical characteristics of coherent graph, the Bayesian classifier with maximum posterior probability is used to divide the coherent map initially. At the same time, the Markov random field is introduced, and the neighborhood information of the coherent graph is used to remove the clutter effectively. The experimental results show that the extracted coherent image characteristics and spatial information can achieve the interference SAR coherent image segmentation. 2. A coherent map migration clustering algorithm for SAR image segmentation is proposed. The Kmeans method is used to divide the target SAR image. According to the initial label, the K-SVD algorithm is used to learn a dictionary for each class of data. In order to reduce the computational complexity and reduce the randomness of the algorithm, according to the clustering centers of each class, a group of samples closest to the center are selected as training data to train the dictionary. In the process of target SAR image segmentation, the separable information of coherent image is transferred to the target SAR image by sparse representation dictionary. Two real interference SAR image experiments show that the segmentation effect is improved obviously. This work is supported by the National Natural Fund "Nos.6100319891472306" and the special fund for basic scientific research business expenses of central colleges and universities (JDYB140508).
【学位授予单位】：西安电子科技大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TN957.52

【相似文献】