【作者】 裴彩红；

【导师】 邹斌；

【作者基本信息】 哈尔滨工业大学 ， 信息与通信工程， 2007， 硕士

【摘要】 POL-SAR属于目前SAR领域和雷达极化领域研究的前沿课题,同时也是研究的热点和难点。在此背景下,开展关于POL-SAR的图像特征提取以及目标分类等方面的研究,对于解决如何优化POL-SAR系统整体性能,提高对POL-SAR图像的解译能力,推动我国POL-SAR系统走向应用具有重要的理论和实际意义。本文研究的是POL-SAR图像特征提取和分类技术,研究的主要内容包括:基于目标分解的POL-SAR图像特征提取,基于灰度共生矩阵的纹理特征提取,以及基于SVM的POL-SAR图像分类。本文首先介绍了POL-SAR系统的历史与发展现状,并针对极化图像目标分解和分类技术的国内外研究现状进行了详细的综述。同时,介绍了SVM以及它在各种不同领域中的发展应用现状,并对SVM理论进行了详细的介绍。接下来,介绍了目标极化散射特性的表示方法,包括极化散射矩阵,协方差矩阵和相干矩阵等,给出了几种基本的极化散射机理。在此基础之上进行了基于目标分解的多种极化散射特征提取,研究了基于灰度共生矩阵的纹理特征提取,进而用ESAR L波段真实数据进行了实验,然后对实验结果做了详尽的分析和比较。最后,着重研究了基于SVM的POL-SAR图像特征选择和分类技术。重点研究了SVM参数的选择,比较了不同的参数和分类精度的关系,提出了具体的分类算法,并用这种算法对POL-SAR图像进行分类。本文还用Cameron分类、H ?α分类和神经网络等方法进行分类,并和SVM分类结果进行了比较,从而验证SVM方法用于POL-SAR图像分类的有效性。更多还原

【Abstract】 POL-SAR is a new kind of radar. Its imaging technique is developed from SAR. Carried forward the merit of SAR, POL-SAR can create high resolution images. Because of the wide and latent applications of POL-SAR in many areas of military, civil and scientific researches, POL-SAR is one of the most active fields in Radar and Remote Sensing nowadays.In this dissertation,the classification of POL-SAR image is studied. Main aspects include scattering feature extraction of POL-SAR image base on target decomposition, texture feature extraction based on gray-level o-occurrence matrix and POL-SAR image classification based on SVM.This dissertation is organized as follows, firstly, an overview of the domain of POL-SAR is provided. Then the development and achievement of POL-SAR decomposition and POL-SAR image classification is provided in detail. SVMand its application in different areas are introduced, followed by an introduction of the basic theory of SVMin chapter 2.Next, the polarimetric scattering characteristics of the target, including polarimetric scattering matrix, covariance matrix, coherence matrix and several polarimetric scattering mechanisms, are introduced. POL-SAR image feature extraction is deep discussed combined with examples. On this basis, a set of feature extraction methods are presented, Corresponding experiments are done using ESAR L-band data. The results are also provided and detailedly analyzed.The last step, feature selection and POL-SAR image classification based on SVM is mainly studied. Parameter selection of SVMis studied. The relationship of different parameters is compared. Then, classification arithmetic is proposed and used for POL-SAR image classification. Another classification experiments are also done. The method we proposed is validated through comparing with the results of other classification methods.更多还原