中国科学院机构知识库网格
Chinese Academy of Sciences Institutional Repositories Grid
基于马尔可夫随机场的遥感图像变化检测关键技术研究

文献类型:学位论文

作者陈克明
学位类别工学博士
答辩日期2010-12-04
授予单位中国科学院研究生院
授予地点中国科学院自动化研究所
导师周志鑫
关键词变化检测 马尔可夫随机场 自适应 多视角协作 机器学习 change detection Markov random field (MRF) adaptive multiview learning machine learning
其他题名change detection of remote sensing images based on Markov Random Fields
学位专业模式识别与智能系统
中文摘要遥感图像变化检测技术在自然灾害监测、国土资源规划管理、军事目标打击评估等众多军民领域都有重要的应用价值。虽然遥感图像变化检测研究已经取得了很大的进步,但仍然处于探索阶段,遥感图像变化检测技术还很不成熟。特别是近年来对遥感图像变化检测应用需求的日益增长,传统的变化检测方法远远不能满足实际的应用需求。因此,有必要对遥感图像变化检测技术做进一步的深入研究。 本文对现有的变化检测方法进行了系统地综述,分析和总结了遥感图像变化检测中存在的问题和困难。针对遥感图像变化检测技术的研究现状和存在的问题,本文以马尔可夫随机场分析方法为理论基础,对遥感图像变化检测的关键技术进行了深入的研究与探讨。本文的主要工作和贡献包含以下几个方面: (1)提出了一种基于马尔可夫随机场融合的无监督合成孔径雷达(synthetic aperture radar (SAR))图像变化检测方法。该方法通过马尔可夫融合策略,有效地缓解了合成孔径雷达图像噪声去除和纹理保护之间的矛盾,既能保持检测结果的平滑性,又能保护图像原有的细节信息,提高检测精度。 (2)提出了一种基于图割理论(graph cuts)的无监督合成孔径雷达图像变化检测方法。该方法把合成孔径雷达图像的变化检测转化为马尔可夫随机场能量函数最小化问题,针对现有马尔可夫随机场能量函数最小化算法易陷于局部极值的缺点,将图割理论应用到基于马尔可夫随机场模型的合成孔径雷达图像变化检测中,实现对能量函数的求解,提高了变化检测的精度。 (3)提出了一种基于自适应马尔可夫随机场的光学图像变化检测方法。针对高分辨率遥感图像丰富的颜色信息和纹理信息,该方法根据不同区域颜色和纹理特征的差异,构造以区域为单位的马尔可夫空间上下文模型,自适应地选取马尔可夫能量函数的邻域阶数和平滑权重系数,实现对输入待检图像的更加精确建模。该方法能较准确地保持检测结果的边缘细节。 (4)提出了一种基于多视角协作策略的遥感图像变化检测方法。在总结了遥感图像变化检测中三种多视角协作情况的基础上,一方面,为了克服传统最大期望(EM)算法容易收敛于局部值的缺点,将变分贝叶斯思想引入到遥感图像变化检测,提出了一种基于概率推荐的变分多视角协作的变化检测框架;另一方面,将高斯过程(Gaussian Process)分类器与co-training算法相结合,扩展传统的co-EM方法到co-GP框架,提出了一种多视角协作(multi-view learning)的高斯过程方法,并应用于高分辨率遥感图像变化检测。 (5)提出了一种多视角协作的马尔可夫随机场变化检测方法。该方法把遥感图像变化检测问题转化为马尔可夫随机场下多个视角的联合能量最小化问题,一方面充分地利用了图像有限的特征,另一方面定量地给出了多视角图像变化检测的计算方法。该方法还第一次将两幅图像的分割问题和变化检测问题统一到了一个框架下。在该框架下,基于两个时相的遥感图像分割问题成为变化检测问题的一个特例。
英文摘要Change detection is one of the most important applications in remote sensing society. It is quite important to military and civil applications, such as the monitoring of natural disasters, the plan and management of land resources, and the evaluation of hitting effect of military targets. Though a great progress has been made in remote sensing change detection filed, it still keeps itself in its infancy and can not meet the actual demand. This is especially true for very high resolution (VHR) image. Because of the complexity of VHR images, the methods for low/moderate resolution images can not be directly applied to VHR image change detection. Therefore, it is necessary to further investigate on the key techniques related to remote sensing image change detection. The dissertation starts with a summarization of existing techniques about remote sensing image change detection, followed by a discussion of the difficulties in remote sensing image change detection from the viewpoint of image processing principle. With the well analysis of the advantages and limits of change detection techniques in existing literature, the dissertation focuses on Markov Random Fields (MRF) based remote sensing image change detection related key techniques investigation. The main contents consist of: (1)A novel unsupervised change detection approach in temporal sets of synthetic aperture radar (SAR) images using Markovian fusion is proposed. This method is capable of resolving the incompatibility of speckle noise removing and spatial information preserving. It can achieve noise removing and spatial information preserving at the same time, which maintains the change results smoothness and improves the detection accuracy. (2)An unsupervised change detection approach for SAR image change detection using graph cuts is proposed. The proposed method treats SAR image change detection in term of energy minimization. To overcome the demerits of traditional energy function minization algorithms, graph cuts algorithm is introduced to SAR image change detection for global optimization, which improves the change detection precision. (3)A novel change detection technique for optical remote sensing imagery under the adaptive MRF framework with combined color and texture features is proposed. Considering the abundant color and texture information contained in VHR images, the proposed approach can adaptively choose the order of neighborhood and the smooth weight coefficient for MRF model. Th...
语种中文
其他标识符200618014628032
源URL[http://ir.ia.ac.cn/handle/173211/6313]  
专题毕业生_博士学位论文
推荐引用方式
GB/T 7714
陈克明. 基于马尔可夫随机场的遥感图像变化检测关键技术研究[D]. 中国科学院自动化研究所. 中国科学院研究生院. 2010.

入库方式: OAI收割

来源:自动化研究所

浏览0
下载0
收藏0
其他版本

除非特别说明,本系统中所有内容都受版权保护,并保留所有权利。