【作者】 徐国华；

【导师】 张保明；

【作者基本信息】 解放军信息工程大学 ， 摄影测量与遥感， 2011， 硕士

【摘要】 遥感影像的变化检测技术在国民经济和国防建设中具有广泛的应用。利用同一地区的多时相遥感影像进行变化检测可以提取、定位该区域在不同时相间的地表变化信息,并且能够定量的对变化信息及变化过程进行比较分析。利用变化检测技术获取的结果可以用于地理数据更新及应用、自然灾害监测评估及预报、战场态势分析以及打击效果评估等多个邻域。本文主要围绕从不同时相遥感影像快速有效提取变化信息展开研究,对面向对象的遥感影像变化检测所涉及的相关理论和方法进行了深入探讨。本文的主要工作和创新点如下:1.对变化检测的现状进行了研究,总结了变化检测方法所存在的问题。深入分析了对象的表达、特性及应用,并归纳和总结了对象分割的方法。将面向对象影像分析方法应用于多时相遥感影像变化检测。2.提出了基于地物空间结构的误匹配剔除方法。针对遥感影像匹配出现的误匹配情况,在顾及变化检测的两幅遥感影像未变化区域具有相同或相似的相对稳定空间结构信息基础上,提出了在待配准影像中通过构建三角形的方式模拟地物空间结构的方法来剔除误匹配点。并通过实验证明了该方法的有效性。3.以对象的提取为目标,重点讨论了将影像划分成内部特性相对均一、相互之间有所差异的影像对象的不同分割方法。研究了均值漂移法和分形网络演化法。将分水岭方法引入到面向对象的变化检测分割中,针对分水岭方法过度分割问题,提出在梯度图像上进行分割处理,再进行滤波然后采用基于局部同质性合并小区域,实验表明过度分割大大减少,较好的实现了对象的提取。4.将变化向量分析法应用于面向对象的变化检测。在对象变化向量分析法中,讨论了利用层间逻辑值对变化强度进行描述的方法。为了快速、有效的获取影像变化信息,采用二维最大类间方差进行变化阈值的自动获取,考虑到二维最大类间方法的速度较慢,提出了一种自动提取阈值的快速计算方法,并通过实验验证了该方法在对象级变化检测中的有效性。论文中对所提出的各种算法利用不同多时相遥感影像进行实验,开发了一个面向对象的变化检测演示验证系统。实验表明采用面向对象的变化检测方法整体上比基于像素的变化检测方法效果要好。更多还原

【Abstract】 Change detection of remote sensing images has been applied extensively in civil and military areas. By analyzing the multi-temporal remote sensing images, change detection can be used to extract and locate the ground change information and it can also be used to analyze the change information and change progress quantitatively. The extracted change information can be used in updating geography date and application, monitoring evaluating and predicting natural disasters, as well as battlefield situation analysis and evaluation of strike effectiveness.This dissertation mainly studies that how do automatic extract changed information from multi-temporal remote sensing images, which discuss deeply some theory and methods on Object-Oriented remote sensing images change detection. The major innovations of this dissertation are listed as follows:1. The development background and status of change detection are discussed, existing problems of change detection approaches and theory are summarized. The expression, attribute and application of objects are deeply analyzed. The segmentation of object are concluded and summarized in the round. Object-oriented image analysis method was applied to multi-temporal remote sensing image change detection.2. Spatial structure is proposed based on the Error matching method. Aimed to the Error matching, taking into account the change detection does not change in two areas of remote sensing images, with the same or similar spatial structure of relative stability based on the information presented in the pending registration. Via constructing a triangle image to simulate the spatial structure of surface features to eliminate false matching points. Experiments show the effectiveness for the method.3. In order to extract object, focusing on dividing the image into internal characteristics relatively homogeneous, the difference between them. A lot image objects are segmented. In this paper, firstly discuss the mean shift method and the fractal net evolution approach. Then watershed method is imported to object-oriented change detection segmentation, Aimed to watershed approach have the immoderacy segmentation problem, Then propose in the gradient image processing, while filter and unite small areas based on local homogeneity. Experiment proved the notable reduction of excessive division and the extraction of object is achieved.4. The change vector analysis is applied to Object-Oriented change detection of remote sensing images. The method of expression variation intension by making use of logicals of layer is discussed of object change vector analysis. In order to obtain the changing intensity of the image rapidly and effectively, the automatic obtain of changes threshold by making use of two-dimensional Otsu. The rapid calculation method is proposed according to the slow speed of two-dimensional Otsu. Experiment proved the feasibility of this method.All the arithmetic are experimented by making use of multi-temporal remote sensing images, and exploit an object-oriented change detection system. Object-oriented approach is better than the pixel-based change detection method.更多还原