遥感图像特征提取与匹配关键技术研究

发布时间：2018-10-29 13:10

【摘要】：卫星遥感技术已成为目前国内外建立空天感知系统的研究重点。作为遥感图像处理中至关重要的一步,遥感图像的配准技术受到了越来越多的关注,其中,遥感图像的特征提取和特征匹配技术是基于特征的遥感图像配准技术的基础,直接影响最终配准的结果。在此背景下,本文重点研究了多源遥感图像配准技术中特征提取和特征匹配的关键技术,主要做了以下几方面研究:针对遥感图像特点,首先对遥感图像中噪声去除和辐射矫正进行了图像预处理分析,并系统地介绍了遥感图像配准基础理论。针对遥感图像特征提取,介绍了互信息特征、SIFT特征和SURF特征的提取原理,并引入信息熵和网格划分的思想,在SURF特征提取前对大幅面遥感图像进行二级网格划分,通过根据信息熵提取特征网格的方式大大减少了特征提取计算量。针对灰度差异较大的多源遥感图像,根据遥感图像边缘轮廓特征,提出了一种基于多项式拟合的形状内容特征提取算法。在对轮廓边缘使用多项式拟合算法提取特征点的基础上,改进了形状内容圆形模板使其对于旋转缩放平移变换均具有良好的不变性。实验证明本文提出算法在速度和鲁棒性上均具有优势。针对遥感图像特征匹配,分别对于SURF特征描述子和改进的形状内容描述子提出了不同的粗匹配算法,在此基础上,提出了将RANSAC算法和互信息相结合的精匹配算法。实验证明,本文方法能够实现多源遥感图像自动精确的配准需要。
[Abstract]:Satellite remote sensing technology has become the focus of research on the establishment of space and space sensing systems at home and abroad. As a very important step in remote sensing image processing, the technology of remote sensing image registration has been paid more and more attention. Among them, the feature extraction and feature matching technology of remote sensing image is the basis of feature based remote sensing image registration technology. Directly affect the result of final registration. Under this background, this paper focuses on the key techniques of feature extraction and feature matching in multi-source remote sensing image registration technology. Firstly, the noise removal and radiation correction in remote sensing image are analyzed, and the basic theory of remote sensing image registration is introduced systematically. Aiming at feature extraction of remote sensing image, the extraction principle of mutual information feature, SIFT feature and SURF feature is introduced, and the idea of information entropy and mesh division is introduced. Before SURF feature extraction, the large format remote sensing image is divided into two levels of grid. The computation of feature extraction is greatly reduced by extracting feature grid according to information entropy. According to the edge contour feature of remote sensing image, a shape content feature extraction algorithm based on polynomial fitting is proposed for multi-source remote sensing image with large gray level difference. On the basis of using polynomial fitting algorithm to extract feature points from contour edge, the circular template of shape content is improved so that it has good invariance for rotation and zoom translation transformation. Experiments show that the proposed algorithm has advantages in speed and robustness. For remote sensing image feature matching, different coarse matching algorithms are proposed for SURF feature descriptors and improved shape content descriptors. Based on this, a precise matching algorithm combining RANSAC algorithm and mutual information is proposed. Experiments show that this method can realize the automatic and accurate registration of multi-source remote sensing images.
【学位授予单位】：南京理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP751

【参考文献】