基于散射特性和空间特性的极化SAR地物分类研究

发布时间：2018-02-24 21:36

本文关键词： 极化SAR 地物分类散射特性多特征集策略空间特性　出处：《西安电子科技大学》2015年硕士论文　论文类型：学位论文

【摘要】：极化合成孔径雷达(Polarimetric Synthetic Aperture Radar,PolSAR)利用四个极化通道可以获得地物目标丰富的极化散射信息,在其众多应用中,极化SAR数据地物分类是最主要的应用领域之一。许多已有的极化SAR数据地物分类方法虽然能够取得不错的分类结果,但是它们都是直接在一组特征集上进行数据分析和分类器设计,且在分类过程中没有考虑极化SAR数据的空间关系。这都可能使分类结果不好,尤其是在分类高分辨率极化SAR数据时。本文主要研究基于散射特性和空间特性的极化SAR地物分类方法,在原有分类方法基础上提出了两种改进方法:1.提出了一种极化SAR数据地物分类方法:基于多特征集的极化SAR地物分类。在该方中采用新的极化特征分析策略——多特征集策略,即对比分析多组非相关极化特征矢量集,在其基础上设计分类器,以改进原有分析一组极化特征集的分类方法。为了验证该策略的有效性,本章改进了原有Wishart迭代分类器——在两组非相关极化特征矢量集上,使用K-MEANS聚类分析,为Wishart分类器选择可分度高的训练样本。高可分度的训练样本不仅加速了Wishart分类器的收敛过程,使其迭代次数大为减少,而且分类器的分类正确率也有所提高,分类后数据的区域一致性更好。真实极化SAR数据实验验证了本章多特征集策略的有效性。2.提出了一种极化SAR数据地物分类方法:基于空间特征和统计特性的极化SAR地物分类。已有的分析极化SAR数据空间关系的地物分类方法在分类前和(或)分类后直接在分类结果中引入分类标号的空间约束,取得了比基于像素的分类方法更好的分类结果。本章从新的角度——在极化特征空间上引入数据的空间约束,设计在分类过程中引入数据空间关系的分类器。为了验证该方法的有效性,本章改进了原有Wishart迭代分类器——通过双边滤波方法在Wishart距离特征上引入极化SAR数据的空间约束,最后分类。改进的Wishart分类器保留了原有分类器的特点,还较好的克服分类器易受相干斑噪声影响的缺点,而且分类后数据的区域一致性更好。真实极化SAR数据实验验证了本章方法的有效性。
[Abstract]:Polarimetric Synthetic Aperture Radarr (PolSAR) can obtain rich polarimetric scattering information from ground objects by using four polarimetric channels, which are widely used in many applications. Ground object classification of polarized SAR data is one of the most important application fields. Many existing methods for ground object classification of polarized SAR data can obtain good classification results. However, they are all designed for data analysis and classifier design directly on a set of feature sets, and the spatial relationship of polarized SAR data is not considered in the classification process, which may result in bad classification results. Especially in the classification of high resolution polarimetric SAR data, this paper mainly studies the polarimetric SAR ground objects classification method based on scattering and spatial characteristics. On the basis of the original classification method, this paper puts forward two improved methods: 1. This paper presents a method of ground objects classification of polarimetric SAR data: poled SAR ground objects classification based on multi-special collection. In this method, a new polarimetric feature analysis strategy, called multi-feature solicitation strategy, is adopted. In order to verify the effectiveness of the strategy, a classifier is designed on the basis of comparative analysis of multiple sets of independent polarization feature vectors to improve the classification method of analyzing a set of polarimetric feature sets. In this chapter, we improve the original Wishart iterative classifier, which uses K-MEANS clustering analysis on two sets of uncorrelated polarization feature vectors. The highly divisible training samples are selected for the Wishart classifier. The convergence of the Wishart classifier is accelerated and the number of iterations is greatly reduced, and the classification accuracy of the classifier is also improved. The experimental results of real polarized SAR data verify the effectiveness of the multi-feature solicitation strategy in this chapter. 2. A method of ground object classification for polarized SAR data is proposed: polarization based on spatial features and statistical characteristics. SAR ground objects classification. The existing methods to analyze the spatial relationship between polarimetric SAR data have introduced the spatial constraints of classification labels directly into the classification results before and / or after classification. A better classification result than the pixel based classification method is obtained. In this chapter, the spatial constraints of the data are introduced into the polarimetric feature space from a new perspective. In order to verify the validity of this method, we improve the original Wishart iterative classifier, which introduces the spatial constraint of polarimetric SAR data on the Wishart distance feature by using the two-sided filtering method. Finally, the improved Wishart classifier preserves the characteristics of the original classifier and overcomes the disadvantages of the classifier which is vulnerable to speckle noise. Moreover, the regional consistency of the classified data is better. The experimental results of real polarimetric SAR data demonstrate the effectiveness of this method.
【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】：2015
【分类号】：TN957.52

【相似文献】