高分辨率PolSAR图像的超像素分割方法研究
发布时间：2018-10-17 12:36
【摘要】：极化合成孔径雷达(PolSAR)通过极化组合天线发射和接收电磁波,从接收电磁波中可以获取地物丰富的散射信息。相比于单极化或者双极化的SAR系统,PolSAR系统能够测量地物目标更多的信息,也能更真实的反映地物目标的情况。随着PolSAR图像分辨率的提高,虽然高分辨率图像可以提供更多的纹理信息、结构信息,但是PolSAR图像的逐像素解译随之出现了一些问题:PolSAR图像中具有严重的相干斑噪声,这就使得图像中很多像素点是类别不确定的噪声点,采用逐像素的方法对其进行解译,就会造成噪声点被错误处理;在高分辨率PolSAR图像中,相邻像素很有可能属于同一种地物,因此这些像素中含有的信息基本相似,如果对这些像素单独处理,则会造成信息冗余。为了解决逐像素解译高分辨率PolSAR图像带来的一些问题,有学者提出了超像素的概念。超像素,是指图像中在形状、颜色、纹理等方面相似度很高的像素点组成的均匀图像块。在高分辨率图像中,数据量很大,像素数有成千上万个,经过超像素分割,后续就变为对几百个超像素处理。超像素的概念是在光学图像中提出的,本文根据PolSAR图像的极化特性,将超像素分割应用在PolSAR图像中,主要研究内容有以下三个方面:首先,研究了PolSAR的基础理论和极化数据的表征方法,研究了多个表征方式之间的相互转换关系,为了从PolSAR图像中提取丰富的极化信息和纹理信息,研究了PolSAR图像的目标分解方法和常用的纹理特征提取方法。其次,研究了在PolSAR图像中的相异性计算方法,根据提取的极化特征和纹理特征,提出利用特征之间的欧式距离计算相异性的思想;研究了超像素分割的基本原理,以及传统的SLIC算法和传统的分水岭算法;结合PolSAR图像的特性,对这两种算法进行改进,使其在PolSAR图像中能够获取边缘贴合度高的超像素。最后,研究了超像素分割结果的评价准则,给出了适合PolSAR图像的评价准则,并利用三幅实验数据进行实验验证,并对结果进行定性和定量评价。
[Abstract]:Polarimetric synthetic Aperture Radar (PolSAR) transmits and receives electromagnetic waves through polarimetric combined antennas, which can obtain abundant scattering information from the received electromagnetic waves. Compared with SAR system with single or double polarization, PolSAR system can measure more information of ground object and reflect the situation of ground object more realistically. With the improvement of the resolution of PolSAR images, although high-resolution images can provide more texture information and structure information, there are some problems in the pixel by pixel interpretation of PolSAR images: there is serious speckle noise in PolSAR images. As a result, many pixels in the image are noise points with uncertain categories, which can be interpreted by the method of pixel by pixel. In high-resolution PolSAR images, adjacent pixels are likely to belong to the same ground object. Therefore, the information contained in these pixels is basically similar, if these pixels are processed separately, the information will be redundant. In order to solve some problems caused by pixel interpretation of high resolution PolSAR images, some scholars have proposed the concept of super pixel. Super pixel is a uniform image block composed of pixels with high similarity in shape, color, texture and so on. In high-resolution images, the amount of data is very large, there are tens of thousands of pixels, after super-pixel segmentation, the subsequent processing becomes hundreds of super-pixels. The concept of super-pixel is proposed in optical image. According to the polarization characteristics of PolSAR image, this paper applies hyperpixel segmentation to PolSAR image. The main research contents are as follows: first, In order to extract rich polarization and texture information from PolSAR images, the basic theory of PolSAR and the representation method of polarization data are studied. The target decomposition and texture feature extraction methods of PolSAR images are studied. Secondly, the different computing method in PolSAR image is studied. According to the extracted polarization feature and texture feature, the idea of using the Euclidean distance between the features to calculate the difference is put forward, and the basic principle of super-pixel segmentation is studied. And the traditional SLIC algorithm and the traditional watershed algorithm, combined with the characteristics of the PolSAR image, the two algorithms are improved to obtain the super-pixel with high edge fit in the PolSAR image. Finally, the evaluation criteria of the super-pixel segmentation results are studied, and the evaluation criteria suitable for PolSAR images are given. The experimental results are verified by using three experimental data, and the results are evaluated qualitatively and quantitatively.
【学位授予单位】：哈尔滨工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TN957.52
本文编号：2276678
[Abstract]:Polarimetric synthetic Aperture Radar (PolSAR) transmits and receives electromagnetic waves through polarimetric combined antennas, which can obtain abundant scattering information from the received electromagnetic waves. Compared with SAR system with single or double polarization, PolSAR system can measure more information of ground object and reflect the situation of ground object more realistically. With the improvement of the resolution of PolSAR images, although high-resolution images can provide more texture information and structure information, there are some problems in the pixel by pixel interpretation of PolSAR images: there is serious speckle noise in PolSAR images. As a result, many pixels in the image are noise points with uncertain categories, which can be interpreted by the method of pixel by pixel. In high-resolution PolSAR images, adjacent pixels are likely to belong to the same ground object. Therefore, the information contained in these pixels is basically similar, if these pixels are processed separately, the information will be redundant. In order to solve some problems caused by pixel interpretation of high resolution PolSAR images, some scholars have proposed the concept of super pixel. Super pixel is a uniform image block composed of pixels with high similarity in shape, color, texture and so on. In high-resolution images, the amount of data is very large, there are tens of thousands of pixels, after super-pixel segmentation, the subsequent processing becomes hundreds of super-pixels. The concept of super-pixel is proposed in optical image. According to the polarization characteristics of PolSAR image, this paper applies hyperpixel segmentation to PolSAR image. The main research contents are as follows: first, In order to extract rich polarization and texture information from PolSAR images, the basic theory of PolSAR and the representation method of polarization data are studied. The target decomposition and texture feature extraction methods of PolSAR images are studied. Secondly, the different computing method in PolSAR image is studied. According to the extracted polarization feature and texture feature, the idea of using the Euclidean distance between the features to calculate the difference is put forward, and the basic principle of super-pixel segmentation is studied. And the traditional SLIC algorithm and the traditional watershed algorithm, combined with the characteristics of the PolSAR image, the two algorithms are improved to obtain the super-pixel with high edge fit in the PolSAR image. Finally, the evaluation criteria of the super-pixel segmentation results are studied, and the evaluation criteria suitable for PolSAR images are given. The experimental results are verified by using three experimental data, and the results are evaluated qualitatively and quantitatively.
【学位授予单位】：哈尔滨工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TN957.52
【参考文献】
相关期刊论文 前6条
1 黄谊;任毅;;基于阈值法和区域生长法的图像分割算法研究[J];电子测试;2012年10期
2 王军辉;李瑞克;刘小洋;;基于改进Sobel算法的实时边缘检测系统[J];计算机与数字工程;2012年06期
3 康牧;许庆功;王宝树;;一种Roberts自适应边缘检测方法[J];西安交通大学学报;2008年10期
4 沈峰亭;魏红;;基于改进Sobel算子的螺纹边缘检测[J];微计算机信息;2008年01期
5 冯建辉;杨玉静;;基于灰度共生矩阵提取纹理特征图像的研究[J];北京测绘;2007年03期
6 王娜,李霞;一种新的改进Canny边缘检测算法[J];深圳大学学报;2005年02期
相关博士学位论文 前2条
1 张腊梅;极化SAR图像人造目标特征提取与检测方法研究[D];哈尔滨工业大学;2010年
2 孙玉宝;图像稀疏表示模型及其在图像处理反问题中的应用[D];南京理工大学;2010年
相关硕士学位论文 前4条
1 韩斌;基于内容的超像素合并及其在图像分割中的应用[D];上海交通大学;2013年
2 刘春燕;图像分割评价方法研究[D];西安电子科技大学;2011年
3 毕芳;基于马尔科夫随机场的纹理图像分类[D];西安理工大学;2010年
4 刘东菊;基于阈值的图像分割算法的研究[D];北京交通大学;2009年
，本文编号：2276678
本文链接：https://www.wllwen.com/kejilunwen/xinxigongchenglunwen/2276678.html
