【作者】 佃袁勇；

【导师】 方圣辉；

【作者基本信息】 武汉大学 ， 摄影测量与遥感， 2005， 硕士

【摘要】 基于遥感影像的变化检测就是从不同时间获取的遥感影像中,定量分析和确定地表变化特征和过程的技术。利用不同时相获取的卫星遥感影像进行变化检测,是开展资源调查、环境监测、基础地理数据库更新等对地观测技术应用中的关键技术,具有广泛的应用领域。迄今为止,众多学者已经提出了很多种关于变化检测的方法,按照是否需要真实的地面数据来分,变化检测方法可以分为:监督法和非监督法。前者是指根据地面真实数据来获取变化区域的训练样区,从而进行变化检测;后者是直接对两个不同时相的数据检测而不需要任何额外的信息。由于地面的真实信息不容易得到,因此非监督的变化检测方法是常用的变化检测方法。非监督的变化检测的流程通常分为三部分:1)预处理即从辐射和几何角度对图像进行处理;2)图像比较,得到变化影像;3)分析变化影像,得到变化特征。 本文的研究重点在分析变化影像。就变化区域的提取问题,展开了研究,主要完成了三个方面的工作: 1)提出了基于贝叶斯理论的最小错误率阈值选择方法。针对差值影像阈值选择困难,本文在总结以往阈值选择方法的基础上,提出了从统计特性的角度采用最大数学期望(EM)方法来获得阈值,并比较了图像服从高斯模型和泛高斯模型时的差异。 2)经过分析发现,单一的阈值方法将每个像素看成独立单元,没有充分考虑相邻像素之间的关系,提取出来的变化区域效果不好。为此,从变化像素的空间邻域考虑,将概率松弛迭代方法和马尔科夫随机场模型运用到变化检测中,取得了较好的效果。 3)针对线状地物的特点,提出了基于边缘特征和灰度的变化检测方法。对于道路、桥梁等线状地物,仅从灰度考虑,进行变化检测,效果不好,根据这些地物自身的特点,我们将边缘特征和灰度特征相结合来考虑变化检测,取得了较好的效果。 实验的结果证明了本文提出的算法能很好的检测出变化区域,并对线状地物的检测有一定的效果。更多还原

【Abstract】 Change Detection in remote sensing imagery is defined as the procedure of quantitatively analyzing and identifying changes occurred on the earth’ s surface from remote sensing imageries acquired at different times. As a key element for many applications of erath observation such as resource inventory, environment monitoring, update of fundamental geographical database, etc,change detection technique is of urgent demands and has great potentiall in scientific applications. So far ,many scients have developed many methods for change detection , and these methods can be classified as superviseed and unsupervised techniques. The former require the availability of a "ground truth" from which to derive a training set containing information about the spectral signatures of changes. The latter performs change detection without any additional information besides the raw images considered. It is obvious that using unsupervised techniques is mandateory in many remote-sensing applications, as suitable ground-truth information is not always available. The change-detection processs performed by such unsupervised techniques is usually divided into three main sequentiall steps:1)pre-precessing, aimed at rendering the two images comparable in both the spatial and spectral domains. 2) image comparison and 3)analysis of the difference image.This paper, we focus on the last step of change-detection process. We have finished three main tasks about the change region extract:Firstly, a method base on Bayes decion for the minimum error is proposed to estabish change threshold. Indor to overcome the diffiult of establish change thresholds in difference image, this paper summarizes the original methods in establishing threshold, and then proposed the Bayes decision for minimum error to establish the threshold which used the Expectation-Maximization algorithm, and compared the difference when the image subject to Gauss and General Gauss Model.Secondly, The top method considered the pixels are all independence and the contextual information is ignored, the result was not good enough. The contextual method is devised for this problem. In this method, probability looseness and markov random field model are used in changedetection .the results improved the accuracy and reliability of the changed area extraction.Thirdly, we proposed line change detection in liner object. Because the above methods are not all consider the line feature of the line object, the results are not so good. We propose a method that considers the edge and gray information to detect the changes, and the results improve that the method is so good.Experimental results demonstrate that the algorithm proprosed in this paper is effective for indentifying and extracting changed areas from difference images.更多还原