基于核变换的遥感影像变化检测

发布时间：2018-06-20 22:51

本文选题：遥感影像变化检测 + 核变换　；参考：《中国矿业大学》2017年硕士论文

【摘要】：遥感对地观测技术凭借其覆盖面广、多时相、周期短、信息丰富等优点,为土地利用现状分析、土地资源管理、灾害预防和应急响应以及城市化进程等各项工程提供更加便捷、有效的服务。针对研究区域变化检测标记样本难以获取、变化信息存在非线性分布等情况,本文引入核思想,从基于像元的遥感影像变化范围和变化类型检测两个目的出发,设计了基于核变换的变化检测方法,主要研究内容如下:(1)在总结核变换的基础理论、核函数的基本属性以及多核构造的必要条件基础上,比较了几类经典的遥感影像变化检测方法,并实现了经典的变化矢量分析法(Change Vector Analysis,CVA),分析其与极化CVA、压缩极化CVA(Compressed CVA,C2VA)间的优劣势,并运用大津阈值法(OTSU)确认变化范围、传统编码法获取变化类型,并以此作为后续实验对比。(2)为了确定变化范围,利用自适应的融合策略融合两时相间经典CVA结果和光谱角、光谱相关性匹配结果,即融合多相似性测度作为最终的变化强度图。采取多目标优化的核K-均值聚类算法代替阈值化过程,在未知数据具体分布形式前提下,增加数据的可分性,在一定程度上提高自动化水平。(3)为了确定变化类型,利用压缩极化CVA获取在极坐标系下能够可视化分析的变化方向信息后,在置信度高的小邻域范围内随机选择训练样本,并运用不同的核函数以不同方式混合的支持向量机(Support Vector Machine,SVM)来确认变化类型。该设计方案与单核SVM相比,节省了人工选择核函数的过程;与传统编码法相比,在精度上有一定提升。(4)在Matlab 2013a平台上设计了一个基于核变换遥感影像变化检测的简易系统,重点开发了三个模块:变化强度融合、变化范围检测和变化类型检测模块。该系统提供便于操作的交互界面,可根据用户选择的具体方法,实现对两时相影像的变化检测。同时,利用两组模拟数据和两组真实影像验证其可行性和有效性,实现了对于不同的数据分布、不同的变化类型影像的变化检测。实验结果表明,相比于传统的变化检测方法,设计的基于核变换的方案表现出较优的检测性能。
[Abstract]:Based on the advantages of wide coverage, multi phase, short period, and rich information, remote sensing technology provides more convenient and effective services for various projects such as land use analysis, land resource management, disaster prevention and emergency response and urbanization process. In this paper, we introduce nuclear thought, design a change detection method based on nuclear transformation based on two purposes, which are based on the range and type of change of remote sensing images. The main research contents are as follows: (1) the basic theory of nuclear transformation, basic attributes of kernel functions and the necessity of multi core construction are summarized. On the basis of conditions, several kinds of classical remote sensing image change detection methods are compared, and the classical variable vector analysis (Change Vector Analysis, CVA) is realized, and the advantages and disadvantages between them and polarization CVA, compressed polarization CVA (Compressed CVA, C2VA) are analyzed. The change range is confirmed by the Otsu threshold method (OTSU), and the change type is obtained by the traditional coding method. And as a follow-up experiment comparison. (2) in order to determine the range of changes, the adaptive fusion strategy is used to fuse the classical CVA results and spectral angles of the two phase, and the spectral correlation matching results, that is, the fusion of multi similarity measure as the final change intensity map. On the premise of knowing the specific distribution of data, increasing the separability of data and improving the level of automation to a certain extent. (3) in order to determine the change type, using the compressed polarization CVA to obtain the change direction information which can be visualized and analyzed in the polar coordinate system, the training samples are randomly selected in the small neighborhood range with high confidence, and the difference is used. The kernel function is Support Vector Machine (SVM) in different ways to confirm the change type. Compared with the mononuclear SVM, this scheme saves the process of artificial selection of the kernel function; compared with the traditional coding method, it has a certain improvement in the precision. (4) a nuclear transform remote sensing image is designed on the Matlab 2013a platform. The simple system of change detection focuses on the development of three modules: change intensity fusion, change range detection and change type detection module. The system provides an easy to operate interface, and can detect the changes of two phase images according to the specific method of user selection. At the same time, two groups of analog data and two groups of real images are used. The feasibility and validity, realizes the different data distribution changes of different types of image detection. The experimental results show that the change detection method compared to the traditional design, the kernel transformation scheme based on the show has a better detection performance.
【学位授予单位】：中国矿业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：P237

【参考文献】