高分辨率极化SAR影像建筑物检测方法研究

发布时间：2018-03-28 08:18

本文选题：极化SAR　切入点：建筑物检测　出处：《电子科技大学》2017年硕士论文

【摘要】：利用极化合成孔径雷达(PolSAR)影像进行地物目标检测和识别是当今极化SAR解译的重要研究课题,从极化SAR影像中检测出建筑物对于土地利用调查、城市规划以及城市变化监测等应用具有重要的理论意义和实用价值。然而由于建筑物具有比较复杂的几何结构分布,并且建筑物的交叉极化散射往往容易导致其被误分为森林等其它地物,因此从极化SAR影像中检测建筑物仍然是一个具有挑战性的课题。本文利用高分辨率极化SAR影像含有的几何特征、纹理特征和极化散射特征等属性来检测建筑物,将建筑物的纹理特征与其极化散射特征相结合,从极化SAR影像中检测出建筑物。本文以L波段星载ALOS-2 PALSAR-2全极化SAR影像和机载E-SAR全极化SAR影像作为实验数据,分别利用极化SAR影像中建筑物的几何特征、纹理特征和极化散射特征来检测建筑物,并且通过加权特征融合方法将纹理特征和极化散射特征等多种特征相结合实现建筑物的检测,主要的研究工作和结论如下:(1)利用基于高分辨率极化SAR影像建筑物几何特征的标记分水岭变换方法检测建筑物。建筑物的边缘轮廓信息是其在极化SAR影像中所呈现出的主要的几何特征,利用基于建筑物几何特征的标记分水岭变换方法从高分辨率极化SAR影像中检测出建筑物,且实验结果表明该方法得到的检测结果对建筑物边缘轮廓信息有较好的保持度。(2)基于极化散射特征对全极化SAR影像进行建筑物检测。本文中建筑物的极化散射特征主要包括极化目标分解参数和圆极化相关系数。主要利用几种常用的极化目标分解来获取建筑物的极化目标分解参数,如Cloude分解、Freeman分解、Yamaguchi分解等。圆极化相关系数对建筑物比较敏感,结合圆极化相关系数和极化目标分解参数,通过Wishart分类器将极化SAR影像分为建筑物和非建筑物,从而区别出极化SAR影像中的建筑物和非建筑物,实现建筑物的检测,实验结果表明结合极化目标分解参数和圆极化相关系数能够有效地从极化SAR影像中检测出建筑物。(3)综合极化SAR影像中建筑物的纹理特征、极化散射特征来检测建筑物。利用加权特征融合的方法将建筑物的纹理特征和极化散射特征等多种特征有效地结合在一起构成建筑物的特征集,再通过SVM分类器将极化SAR影像中的地物目标分为建筑物和非建筑物,从而区别出极化SAR影像中的建筑物和非建筑物,且得到建筑物的整体检测结果较好,还对各检测结果进行分析和评价,实验结果表明基于多特征融合的建筑物检测结果在检出率和正确率等评价指标上都有所提高,因此综合纹理特征和极化散射特征对建筑物检测有较好的影响。
[Abstract]:The detection and recognition of ground objects using polarimetric synthetic aperture radar (SAR) images is an important research topic in the interpretation of polarimetric SAR. The land use survey of buildings has been detected from the polarimetric SAR images. The application of urban planning and urban change monitoring has important theoretical significance and practical value. However, because of the complex geometric structure distribution of buildings, And the cross-polarization scattering of buildings often leads them to be misclassified into other ground objects such as forests. Therefore, it is still a challenging task to detect buildings from polarized SAR images. In this paper, the geometric features, texture features and polarimetric scattering features of high resolution polarimetric SAR images are used to detect buildings. The texture features of buildings are combined with their polarimetric scattering features to detect buildings from polarized SAR images. In this paper, the L-band space-borne ALOS-2 PALSAR-2 fully polarized SAR images and airborne E-SAR all-polarized SAR images are used as experimental data. The geometric features, texture features and polarimetric scattering features of buildings in polarized SAR images are used to detect buildings, respectively. And the texture feature and polarization scattering feature are combined by weighted feature fusion method to realize building detection. The main research work and conclusions are as follows: 1) using the method of tagged watershed transform based on high resolution polarimetric SAR image to detect the building. The contour information of the building is presented in the polarized SAR image. The main geometric features that appear, The building is detected from high resolution polarized SAR images by using the marked watershed transform method based on the geometric features of buildings. The experimental results show that the proposed method has a good preserving degree for building edge profile information. (2) based on the polarimetric scattering characteristics, the building detection of fully polarized SAR images is carried out. In this paper, the polarimetric scattering characteristics of buildings are presented. The characteristics mainly include polarization target decomposition parameters and circular polarization correlation coefficients. Several commonly used polarization target decomposition parameters are mainly used to obtain the polarization target decomposition parameters of buildings. For example, Cloude decomposition Freeman decomposition and Yamaguchi decomposition. Circular polarization correlation coefficient is sensitive to buildings. Combined with circular polarization correlation coefficient and polarimetric target decomposition parameters, polarimetric SAR images are divided into buildings and non-buildings by Wishart classifier. In order to distinguish the building from the non-building in the polarized SAR image, the detection of the building can be realized. The experimental results show that the texture features of buildings in polarimetric SAR images can be effectively detected from polarized SAR images by combining polarization target decomposition parameters with circular polarization correlation coefficients. Using the weighted feature fusion method, the texture feature and the polarization scattering feature of the building are effectively combined to form the building feature set. Then the ground objects in polarized SAR images are divided into buildings and non-buildings by SVM classifier, and the buildings in polarized SAR images are distinguished from non-buildings, and the overall detection results of buildings are obtained. The experimental results show that the detection results of buildings based on multi-feature fusion have been improved in terms of detection rate and accuracy. Therefore, the integrated texture feature and polarization scattering feature have a good effect on building detection.
【学位授予单位】：电子科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TN957.52

【参考文献】