多特征融合的遥感影像变化检测方法研究

发布时间：2018-03-02 00:26

本文关键词： 高分辨率影像面向对象影像配准变化检测多特征融合　出处：《中国矿业大学》2017年硕士论文　论文类型：学位论文

【摘要】：遥感影像对地观测技术具有覆盖面积大、周期性短、信息广等优点,基于遥感影像变化检测能够连续观测到地物变化信息,并且能够定量的对变化信息及变化过程进行比较分析。利用变化检测技术能够快速获取城镇扩张、自然灾害评估、地理国情监测以及战场态势分析等结果,使得高分辨率遥感影像变化检测研究成为的热点之一。现有的高分辨率遥感影像变化检测算法(单一特征方法)存在着不足以及适用范围限制,结合多种特征可以有效利用每种特征的优势,提高算法的检测精度,为变化检测提供新思路。本文针对高分辨率遥感影像自身的特点,对面向对象的变化检测技术进行系统性研究,为解决现有方法存在的问题,提出相应的改进方法。论文的主要工作如下:(1)在系统总结和分析高分辨率遥感影像变化检测的国内外现状和存在问题的基础上,详细介绍了变化检测的技术路线,根据处理单位的不同,将高分辨率遥感影像变化检测分为基于像素和面向对象的变化检测,并对两种方法进行了详细论述。(2)针对相似性场景中遥感影像匹配错误率高的问题,提出的结合全局信息的SIFT特征匹配算法具有以下优点:提出新的DAISY描述子主方向分配方法,变换后的DAISY描述子具有旋转不变性;在DAISY描述子中引入全局纹理信息描述子,当影像存在局部相似区域,借助全局纹理信息来对其进行描述,从而降低了由于局部区域相似而造成的错误匹配的概率。实验结果表明,该算法对较小角度的旋转、光谱变化较大等具有很强的鲁棒性。(3)针对传统的变化检测算法主要依赖像斑的光谱信息,未能有效的利用影像多特征检测优势的问题。基于面向对象的思想,提出一种多特征融合的遥感影像变化检测算法。首先,以多尺度分割的影像对象为基础,计算各对象的颜色直方图和边缘直线梯度直方图;然后,利用EMD算子统计不同时期对象之间的颜色距离和边缘直线特征距离,采用自适应加权的方法将颜色特征距离和边缘直线距离组合构建对象的异质性;最后,采用直方图曲率分析获得像斑的变化检测结果。实验结果表明,该方法能够充分融合颜色和边缘直线特征,提高变化检测的精度。
[Abstract]:Remote sensing image Earth observation technology has the advantages of large coverage, short periodicity, wide information, etc. Based on remote sensing image change detection can continuously observe the ground object change information. And the change information and process can be compared and analyzed quantitatively. By using change detection technology, the results of urban expansion, natural disaster assessment, geographical situation monitoring and battlefield situation analysis can be obtained quickly. It makes the research of high-resolution remote sensing image change detection become one of the hot spots. The existing high-resolution remote sensing image change detection algorithm (single feature method) has some shortcomings and its application scope is limited. Combined with various features, the advantages of each feature can be effectively utilized, the detection accuracy of the algorithm can be improved, and a new idea for change detection can be provided. This paper aims at the characteristics of high resolution remote sensing image. In order to solve the problems existing in the existing methods, the object oriented change detection technology is studied systematically. The main work of this paper is as follows: 1) on the basis of systematically summarizing and analyzing the present situation and existing problems of high-resolution remote sensing image change detection at home and abroad, the technical route of change detection is introduced in detail. According to the difference of processing units, the change detection of high resolution remote sensing image is divided into pixel and object oriented change detection, and the two methods are discussed in detail. The proposed SIFT feature matching algorithm combining global information has the following advantages: a new DAISY descriptor principal direction assignment method is proposed, the transformed DAISY descriptor is rotation-invariant, and a global texture information descriptor is introduced into the DAISY descriptor. When there is a local similar region in the image, the global texture information is used to describe it, thus reducing the probability of the mismatch caused by the local similarity. The experimental results show that the algorithm rotates at a smaller angle. For the traditional change detection algorithms mainly rely on the spectral information of image spots, they can not effectively utilize the advantages of image multi-feature detection. A multi-feature fusion algorithm for remote sensing image change detection is proposed. Firstly, the color histogram and edge linear gradient histogram of each object are calculated based on the multi-scale segmentation image object. The EMD operator is used to calculate the color distance and edge line feature distance between objects in different periods, and the adaptive weighting method is used to combine the color feature distance and edge line distance to construct the heterogeneity of the object. Histogram curvature analysis is used to obtain the change detection results of image spots. The experimental results show that the method can fully integrate color and edge straight line features and improve the accuracy of change detection.
【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;TP751

【参考文献】