多时相多传感器遥感影像配准研究

发布时间：2018-04-13 06:53

本文选题：影像配准 + 特征点提取　；参考：《辽宁工程技术大学》2013年硕士论文

【摘要】：随着新型传感器的不断涌现和遥感技术的不断发展，，人们获取遥感影像的种类和手段不断增多、能力不断提高。为不同时相影像间的变化检测、图像融合、目标识别等多源图像的处理工作提供海量的数据基础。在进行这些工作之前，必须进行影像间的配准，而配准精度的高低往往直接决定着后续应用成果的好坏。因此，本文对多时相多传感器遥感影像之间配准技术的研究就显得十分必要。首先，本文对现在的国内外配准技术的研究概况做了简单扼要的概括。回顾配准技术的理论基础知识，并按配准分类和配准后影像应用分类介绍进一步说明配准方法及其应用现状，并分析出当前配准技术所面临的一些问题，明确了本文研究的目的和意义。其次，本文着重介绍了基于特征点的遥感影像配准的关键技术，分别对影像的预处理、特征点提取及描述、特征匹配、空间变换矩阵的估算、重采样和插值这五步配准流程进行了详细介绍。并通过实验对三种基于SIFT、SURF、Harris-Laplace三种算法的配准方法进行匹配能力、定位精度和时间上的对比，折中选择一种方法作为改进算法的基础。最后，对多时相多传感器大影像现在存在的系统提取特征点速度缓慢、提取特征信息不完全、精度难以保证等问题，提出分块提取特征点、利用最佳匹配点建立TIN三角网结合小面元纠正获得高精度、快速的多时相大影像配准方法，并且通过实验验证了这种方法的可行性。
[Abstract]:With the continuous emergence of new sensors and the development of remote sensing technology, the types and means of obtaining remote sensing images are increasing and the ability of obtaining remote sensing images is increasing.It provides massive data base for multi-source image processing such as change detection, image fusion, target recognition and so on.Before these work, the registration between images must be carried out, and the level of registration accuracy often directly determines the quality of subsequent applications.Therefore, it is necessary to study the registration technology of multi-time and multi-sensor remote sensing images in this paper.Firstly, this paper makes a brief summary of the research of registration technology at home and abroad.Reviewing the basic theoretical knowledge of registration technology, and introducing the registration method and its application status according to the classification of registration and the classification of image application after registration, and analyzing some problems faced by the current registration technology,The purpose and significance of this paper are clarified.Secondly, this paper introduces the key techniques of remote sensing image registration based on feature points, including image preprocessing, feature extraction and description, feature matching, estimation of spatial transformation matrix, etc.The five-step registration process of resampling and interpolation is introduced in detail.Three registration methods based on sift SURFN Harris-Laplace algorithm are compared in matching ability, location accuracy and time, and a compromise method is chosen as the basis of the improved algorithm.Finally, to solve the problems such as the slow speed of feature point extraction, the incomplete feature information extraction and the difficulty of guaranteeing the accuracy of the system, the feature points are extracted in blocks.The best matching point is used to establish TIN triangle network combined with small area correction to obtain high precision and fast multi-phase large image registration method. The feasibility of this method is verified by experiments.
【学位授予单位】：辽宁工程技术大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：P237

【参考文献】