基于纹理和极化特征的SAR影像分类

发布时间：2018-05-28 03:42

本文选题：极化SAR + 纹理特征　；参考：《辽宁工程技术大学》2013年硕士论文

【摘要】：极化SAR具有全天时、全天侯工作特点,同时SAR影像具有丰富的极化信息和纹理特征,被广泛应用于民用及军事领域。SAR影像的解译是它在诸多领域的应用基础,其涉及内容较多,其中极化SAR的分类是关键步骤之一。选取合适的SAR影像特征是获得高精度分类结果的关键,往往一类影像特征仅能反映地物的一方面信息,而多种特征的结合更能突出地物之间的细微区别,更有利于获取准确的分类结果。鉴于SAR影像的纹理特征和极化特征在表现地物特征方面的互异和互补性,本文研究结合纹理特征和极化特征进行SAR影像分类的方法,弥补用单一特征进行影像分类的不足和应用限制。主要工作和贡献体现在以下几个方面： 1)深入研究了SAR影像的纹理特征提取。从灰度共生矩阵和框架小波变换两个方面研究SAR影像纹理特征的提取方法,并对特征提取过程中的参数问题进行了一定分析。应用均值、直方图、相关系数、类分离度方法分析了纹理特征对地物的区分能力。 2)对典型地物的散射机制进行了深入的分析。从极化分解、极化合成、协方差矩阵元素三个方面提取SAR影像的极化参数,对典型地物在极化参数上的不同表现进行定性、定量的分析。 3)总结了基于支持向量机的目标识别原理,给出基于支持向量机的SAR影像多类分类问题的处理方法。 4)研究了基于SVM的全极化SAR影像分类方法。从区分地物的角度出发选择合适的纹理和极化特征,进行了根河地区的影像分类实验,并对分类结果进行精度评价和对比分析,具体分析了GLCM纹理和小波纹理对影像分类精度的影响。
[Abstract]:SAR image is widely used in civil and military fields . SAR image is one of the key steps to obtain high - precision classification results .

In view of the differences and complementarity of SAR image ' s texture and polarization characteristics , this paper studies the method of SAR image classification based on texture features and polarization characteristics , and makes up for the shortage and application limitation of image classification with single feature .

1 ) The texture feature extraction of SAR image is studied deeply . The extraction method of SAR image texture feature is studied from two aspects of gray - scale co - occurrence matrix and frame wavelet transform . The parameter problem in feature extraction is analyzed .

The polarization parameters of SAR image are extracted from three aspects of polarization decomposition , polarization synthesis and covariance matrix element .

3 ) The principle of target recognition based on support vector machine is summarized , and the processing method of multi - class classification problem of SAR image based on support vector machine is given .

4 ) The classification method of polarimetric SAR image based on SVM is studied . The proper texture and polarization characteristics are selected from the angle of distinguishing feature , image classification experiment is carried out in the root river region , and precision evaluation and contrast analysis are carried out on the classification result , and the influence of GLCM texture and wavelet texture on image classification accuracy is analyzed .
【学位授予单位】：辽宁工程技术大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：P237

【参考文献】