SAR图像分割中的变分问题研究
发布时间：2019-05-11 02:36
【摘要】：合成孔径雷达(Synthetic Aperture Radar,SAR)图像分割,是SAR图像处理中的基础且关键步骤。由于相干斑噪声的影响,对SAR图像的分割,需要根据SAR图像所特有的特征来完成。基于变分理论的SAR图像分割,可以根据不同的SAR图像特征建立不同的能量泛函,并利用变分法最小化能量泛函,实现SAR图像的分割。本文就变分SAR图像分割技术展开了研究,主要工作如下:(1)研究了利用边界信息的GAC模型和利用区域信息的CV模型,并结合边界信息和区域信息,建立了多区域的SAR图像分割模型。针对水平集方法数值求解耗时的问题,在Potts模型的基础上推导出了平滑对偶模型,给出了一种基于对偶的快速算法。在SAR图像分割的速度和精确度上,与水平集方法进行了对比,用实验验证了对偶算法的有效性和快速性。(2)分析了均匀SAR图像的Gamma统计分布特征,以及非均匀SAR图像的纹理特征。利用灰度共生矩阵(GLCM)提取SAR图像的纹理特征,得到纹理特征向量,与Gamma分布统计特征进行结合,并利用Potts模型建立了能量泛函。将SAR图像分割结果,与单独使用统计特征或者纹理特征进行对比,特征结合得到的分割结果更加精确。(3)针对极化SAR图像特有的性质,利用极化相干矩阵的复Wishart分布和Potts模型建立能量泛函。对偶算法进行最小化时,在H??分类的基础上,利用复Wishart分布实现了极化SAR图像的自动初始化,可以自动确定初始化曲线和分类数目,考虑了极化SAR图像的统计特征和散射特征。与随机初始化和人工初始化的极化SAR图像分割结果进行了对比,自动初始化分割结果更符合真实的地物信息。
[Abstract]:Synthetic Aperture Radar (Synthetic Aperture Radar,SAR) image segmentation is the basis and key step in SAR image processing. Due to the influence of speckle noise, the segmentation of SAR image needs to be completed according to the characteristics of SAR image. The SAR image segmentation based on variation theory can establish different energy Functionals according to different SAR image features, and use the variation method to minimize the energy Functionals to realize the segmentation of SAR images. In this paper, the variation SAR image segmentation technology is studied, the main work is as follows: (1) the GAC model using boundary information and the CV model using region information are studied, and the boundary information and region information are combined. A multi-region SAR image segmentation model is established. In order to solve the time-consuming problem of level set method, a smooth dual model is derived on the basis of Potts model, and a fast algorithm based on duality is given. The speed and accuracy of SAR image segmentation are compared with the level set method, and the effectiveness and rapidity of the dual algorithm are verified by experiments. (2) the Gamma statistical distribution characteristics of uniform SAR images are analyzed. And the texture features of non-uniform SAR images. The gray co-occurrence matrix (GLCM) is used to extract the texture features of SAR images, and the texture feature vectors are obtained, which are combined with the statistical features of Gamma distribution, and the energy functional is established by using Potts model. The segmentation results of SAR images are compared with those of statistical features or texture features alone, and the segmentation results obtained by the combination of features are more accurate. (3) aiming at the unique properties of polarized SAR images, The energy functional is established by using the complex Wishart distribution of polarization coherence matrix and Potts model. When the dual algorithm is minimized, in H? On the basis of classification, the automatic initialization of polarized SAR image is realized by using complex Wishart distribution, and the initialization curve and classification number can be determined automatically, and the statistical and scattering characteristics of polarized SAR image are considered. Compared with the polarization SAR image segmentation results of random initialization and manual initialization, the automatic initialization segmentation results are more in line with the real ground object information.
【学位授予单位】：电子科技大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TN957.52
本文编号：2474183
[Abstract]:Synthetic Aperture Radar (Synthetic Aperture Radar,SAR) image segmentation is the basis and key step in SAR image processing. Due to the influence of speckle noise, the segmentation of SAR image needs to be completed according to the characteristics of SAR image. The SAR image segmentation based on variation theory can establish different energy Functionals according to different SAR image features, and use the variation method to minimize the energy Functionals to realize the segmentation of SAR images. In this paper, the variation SAR image segmentation technology is studied, the main work is as follows: (1) the GAC model using boundary information and the CV model using region information are studied, and the boundary information and region information are combined. A multi-region SAR image segmentation model is established. In order to solve the time-consuming problem of level set method, a smooth dual model is derived on the basis of Potts model, and a fast algorithm based on duality is given. The speed and accuracy of SAR image segmentation are compared with the level set method, and the effectiveness and rapidity of the dual algorithm are verified by experiments. (2) the Gamma statistical distribution characteristics of uniform SAR images are analyzed. And the texture features of non-uniform SAR images. The gray co-occurrence matrix (GLCM) is used to extract the texture features of SAR images, and the texture feature vectors are obtained, which are combined with the statistical features of Gamma distribution, and the energy functional is established by using Potts model. The segmentation results of SAR images are compared with those of statistical features or texture features alone, and the segmentation results obtained by the combination of features are more accurate. (3) aiming at the unique properties of polarized SAR images, The energy functional is established by using the complex Wishart distribution of polarization coherence matrix and Potts model. When the dual algorithm is minimized, in H? On the basis of classification, the automatic initialization of polarized SAR image is realized by using complex Wishart distribution, and the initialization curve and classification number can be determined automatically, and the statistical and scattering characteristics of polarized SAR image are considered. Compared with the polarization SAR image segmentation results of random initialization and manual initialization, the automatic initialization segmentation results are more in line with the real ground object information.
【学位授予单位】：电子科技大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TN957.52
【参考文献】
相关期刊论文 前1条
1 东野长磊;郑永果;姜东焕;张彬;;基于全局极小解Chan-Vese模型的SAR图像分割[J];计算机工程与设计;2012年11期
，本文编号：2474183
本文链接：https://www.wllwen.com/kejilunwen/wltx/2474183.html
