基于非局部小波和水平集的SAR图像变化检测
发布时间：2019-03-16 11:16
【摘要】：变化检测问题属于图像处理领域,通常是指“根据不同时间的多次观测来确定一个物体的状态变化或确定某种现象的变化的过程”。随着遥感技术的发展,合成孔径雷达(Synthetic Aperture Radar,SAR)图像成为了图像变化检测问题中的主要的数据来源。SAR图像的获得具有全天候、全天时的特点,合成孔径雷达在成像过程中对地物有一定的穿透能力,且不受大气、气候等随机因素的影响,具有不可比拟的优点。国内外众多学者对SAR图像的变化检测问题进行了大量的研究,变化检测结果的精度也在不断地提高。无监督变化检测算法为最常使用的变化检测算法,该类算法的主要步骤为SAR图像的预处理、差异图的构造和差异图的分析,本文的研究重点为SAR图像变化检测中的差异图构造和差异图分析的问题。本文在两方面对SAR图像变化检测技术进行了提高,具体如下所述:1.对于差异图的构造,提出了一种基于非局部小波信息的SAR图像变化检测方法。这种方法提出对同一地区的两幅遥感图像先用简单的代数方法产生差异图,再对差异图进行小波分解,对于高频部分,使用基于非局部均值的方法进行去噪,将高频图像的每一个像素点结合邻域信息转变为向量,对图像进行升维,再利用高斯核函数判断全局信息对于这个点去噪的加权系数,然后求加权均值得到这个点的真实灰度值,对高频图像的每一个点都进行相同的操作,最后进行小波逆变换,得到最终的差异图。在高频部分使用基于非局部均值的去噪方法,既能有效地保留图像的结构信息,又能去除噪声,实现了保留细节信息和去除噪声的平衡,将提出的方法与传统的基于代数的方法和图像融合的算法对比,实验结果表明,提出的方法无论是在视觉上还是定量的评价指标上都能取得较好的结果;2.对于差异图分析,提出了一种基于水平集的动态轮廓模型,这种模型的提出是基于局部模糊C均值聚类算法,对局部模糊C均值聚类算法的目标函数中加入水平集函数构成局部能量函数,使用高斯核函数使得局部信息对于目标能量方程的贡献是可控的,在目标能量方程中加入使得水平集函数能够保持良好的形状和约束零水平集曲线光滑演化的正则项,与全局能量项一起构成新的目标能量方程,最小化目标能量方程,得到目标区域的轮廓曲线。实验证明,相比于之前的经典水平集方法和局部模糊C均值聚类算法,所提出的方法在差异图分割方面能够取得较好的结果。
[Abstract]:The problem of change detection belongs to the field of image processing. It usually refers to the process of determining the state change of an object or determining the change of a phenomenon according to many observations at different times. With the development of remote sensing technology, synthetic Aperture Radar (Synthetic Aperture Radar,SAR) images have become the main data source in image change detection. Synthetic Aperture Radar (SAR) has a certain penetrating ability to surface objects in the imaging process, and is not affected by random factors such as atmosphere and climate, so it has unparalleled advantages. Many scholars at home and abroad have done a lot of research on the change detection of SAR images, and the accuracy of the change detection results is constantly improving. Unsupervised change detection algorithm is the most commonly used change detection algorithm. The main steps of this kind of algorithm are pre-processing of SAR image, construction of difference graph and analysis of difference graph. This paper focuses on the construction of difference graph and the analysis of difference graph in SAR image change detection. In this paper, the SAR image change detection technology has been improved in two aspects, as follows: 1. For the construction of difference graph, a SAR image change detection method based on nonlocal wavelet information is proposed. This method proposes that two remote sensing images in the same area are generated by simple algebraic method and then decomposed by wavelet transform. For the high frequency part, the method based on non-local mean is used to Denoise the image, and the difference image is decomposed by wavelet transform, and the method based on non-local mean is used to remove the noise. Every pixel of high frequency image is transformed into vector combined with neighborhood information, the dimension of the image is raised, and then the weighted coefficient of global information for denoising of this point is judged by Gao Si kernel function. Then the real gray value of the point is obtained by the weighted mean, and the same operation is performed on each point of the high-frequency image. Finally, the inverse wavelet transform is carried out to obtain the final difference graph. In the high frequency part, the denoising method based on the non-local mean can not only preserve the structure information of the image effectively, but also remove the noise, and achieve the balance between preserving the detail information and removing the noise. The proposed method is compared with the traditional algebra-based method and the image fusion algorithm. The experimental results show that the proposed method can achieve good results both visually and quantitatively. 2. For difference graph analysis, a dynamic contour model based on level set is proposed, which is based on local fuzzy C-means clustering algorithm. The local energy function is formed by adding the level set function to the objective function of the local fuzzy C-means clustering algorithm. The contribution of the local information to the target energy equation is controllable by using the Gao Si kernel function. A regular term is added to the objective energy equation, which makes the level set function maintain a good shape and constrains the smooth evolution of the zero level set curve. Together with the global energy term, a new objective energy equation is formed to minimize the objective energy equation. The contour curve of the target area is obtained. The experimental results show that compared with the classical level set method and the local fuzzy C-means clustering algorithm, the proposed method can achieve better results in the segmentation of difference graph than the classical level set method and the local fuzzy C-means clustering algorithm.
【学位授予单位】：西安电子科技大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TN957.52
[Abstract]:The problem of change detection belongs to the field of image processing. It usually refers to the process of determining the state change of an object or determining the change of a phenomenon according to many observations at different times. With the development of remote sensing technology, synthetic Aperture Radar (Synthetic Aperture Radar,SAR) images have become the main data source in image change detection. Synthetic Aperture Radar (SAR) has a certain penetrating ability to surface objects in the imaging process, and is not affected by random factors such as atmosphere and climate, so it has unparalleled advantages. Many scholars at home and abroad have done a lot of research on the change detection of SAR images, and the accuracy of the change detection results is constantly improving. Unsupervised change detection algorithm is the most commonly used change detection algorithm. The main steps of this kind of algorithm are pre-processing of SAR image, construction of difference graph and analysis of difference graph. This paper focuses on the construction of difference graph and the analysis of difference graph in SAR image change detection. In this paper, the SAR image change detection technology has been improved in two aspects, as follows: 1. For the construction of difference graph, a SAR image change detection method based on nonlocal wavelet information is proposed. This method proposes that two remote sensing images in the same area are generated by simple algebraic method and then decomposed by wavelet transform. For the high frequency part, the method based on non-local mean is used to Denoise the image, and the difference image is decomposed by wavelet transform, and the method based on non-local mean is used to remove the noise. Every pixel of high frequency image is transformed into vector combined with neighborhood information, the dimension of the image is raised, and then the weighted coefficient of global information for denoising of this point is judged by Gao Si kernel function. Then the real gray value of the point is obtained by the weighted mean, and the same operation is performed on each point of the high-frequency image. Finally, the inverse wavelet transform is carried out to obtain the final difference graph. In the high frequency part, the denoising method based on the non-local mean can not only preserve the structure information of the image effectively, but also remove the noise, and achieve the balance between preserving the detail information and removing the noise. The proposed method is compared with the traditional algebra-based method and the image fusion algorithm. The experimental results show that the proposed method can achieve good results both visually and quantitatively. 2. For difference graph analysis, a dynamic contour model based on level set is proposed, which is based on local fuzzy C-means clustering algorithm. The local energy function is formed by adding the level set function to the objective function of the local fuzzy C-means clustering algorithm. The contribution of the local information to the target energy equation is controllable by using the Gao Si kernel function. A regular term is added to the objective energy equation, which makes the level set function maintain a good shape and constrains the smooth evolution of the zero level set curve. Together with the global energy term, a new objective energy equation is formed to minimize the objective energy equation. The contour curve of the target area is obtained. The experimental results show that compared with the classical level set method and the local fuzzy C-means clustering algorithm, the proposed method can achieve better results in the segmentation of difference graph than the classical level set method and the local fuzzy C-means clustering algorithm.
【学位授予单位】：西安电子科技大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TN957.52
【相似文献】
相关期刊论文 前10条
1 沈国键，夏启明，，李瑛;基于知识自适应图像平滑[J];武汉测绘科技大学学报;1995年02期
2 林方特;;几种常见图像平滑技术的研究[J];才智;2009年04期
3 杨朝辉;张大鹏;李乃民;;人体生理图像与病理舌纹图像[J];哈尔滨工业大学学报;2009年12期
4 郭琦;;图像平滑的2维小波插值方法[J];中国图象图形学报;2010年10期
5 杨平先,孙兴波;基于粗集方法改进的图像平滑[J];电讯技术;2003年03期
6 王洪亮,曹苏明,刘国平,吴建华;改进的梯度倒数加权算法在图像平滑中的应用[J];红外技术;2003年04期
7 郭海涛,田坦,王连玉,闫宏生;利用粗集理论的声呐图像平滑[J];海洋技术;2005年02期
8 吕振肃,魏弘博,刘R
本文编号：2441225
本文链接：https://www.wllwen.com/kejilunwen/wltx/2441225.html
