基于特征统计的极化SAR图像分类

发布时间：2018-04-17 20:33

本文选题：极化SAR分类 + 联合统计　；参考：《西安电子科技大学》2014年硕士论文

【摘要】：极化合成孔径雷达(Polarimetrie Synthetic Aperture Radar,POLSAR)图像描述的是雷达系统发出的电磁波与地物目标之间的相互关系,它主要呈现了地物目标的分布特性。因此,利用极化SAR图像数据的极化散射信息可以实现地物目标的分类和识别。然而,在极化SAR成像的过程中,电磁波之间的相互干扰以及复杂的地物分布使得复杂场景广泛存在于极化SAR图像中;复杂场景问题不仅影响地物的检测和识别,而且是极化SAR技术定量化发展的重要阻碍。因此,本文主要对复杂场景分类进行研究,主要工作如下:(1)提出了一种基于特征值联合统计的极化SAR图像相似场景分类方法,采用体散射功率和最大特征值的联合高斯模型对数据的分布特性进行了估计,在统计的框架下使用贝叶斯分类器实现了极化SAR图像的相似场景分类;该方法基于马尔科夫随机场(Markov Random Field,MRF)模型对图像的局部纹理信息进行了充分的利用,提高了算法的分类精度和分类结果的区域一致性。(2)提出了一种基于局部高斯模型的极化SAR图像异质场景分类方法,在初始分类的基础上采用局部统计方法对极化SAR图像中的异质场景进行了度量;通过设置自适应的阈值实现了极化SAR图像的异质场景分类。该方法主要采用了局部邻域中主导目标的分布密度表征了异质场景的多样性和易变性。(3)提出了一种基于证据理论的极化SAR图像异质场景分类与识别方法,将极化SAR图像的异质场景作为不确定性问题来考虑,在证据理论的基础上,采用最大特征值的欧氏距离作为信任函数来度量目标点对各种属性的信任度,最后根据最大信任度规则实现了极化SAR图像的异质场景分类与识别。该方法具有较强的理论基础,不仅可以对异质场景进行分类并且可以对异质场景中混杂分布的多种地物目标进行识别。
[Abstract]:Polarimetrie Synthetic Aperture POLSAR (Polarimetrie Aperture POLSAR) images describe the relationship between electromagnetic waves emitted by radar systems and ground objects, which mainly show the distribution characteristics of ground objects.Therefore, the polarimetric scattering information of polarimetric SAR images can be used to classify and recognize ground objects.However, in the process of polarimetric SAR imaging, the interaction between electromagnetic waves and the distribution of complex ground objects make complex scenes widely exist in polarized SAR images, and complex scene problems not only affect the detection and recognition of ground objects.Moreover, it is an important obstacle to the quantitative development of polarized SAR technology.Therefore, this paper mainly studies the classification of complex scenes. The main work is as follows: 1) A method of similar scene classification for polarimetric SAR images based on joint eigenvalue statistics is proposed.The distribution characteristics of the data are estimated by using the combined Gao Si model of volume scattering power and maximum eigenvalue, and the similar scene classification of polarimetric SAR images is realized by using Bayesian classifier under the statistical framework.Based on Markov random field Random Random model, this method makes full use of the local texture information of the image.The classification accuracy of the algorithm and the regional consistency of the classification results are improved. (2) A heterogeneous scene classification method for polarized SAR images based on local Gao Si model is proposed.Based on the initial classification, the heterogeneous scenes in polarized SAR images are measured by local statistics, and the heterogeneous scene classification of polarized SAR images is realized by setting an adaptive threshold.In this method, the distribution density of dominant targets in the local neighborhood is used to characterize the diversity and variability of heterogeneous scenes. A method for heterogeneous scene classification and recognition in polarized SAR images based on evidence theory is proposed.The heterogeneous scene of polarimetric SAR images is considered as an uncertain problem. Based on the evidence theory, the Euclidean distance of the maximum eigenvalue is used as a trust function to measure the trust degree of the target point to various attributes.Finally, the heterogeneous scene classification and recognition of polarimetric SAR images are realized according to the maximum trust rule.This method has a strong theoretical foundation, and it can not only classify heterogeneous scenes but also recognize a variety of ground objects with mixed distribution in heterogeneous scenes.
【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】