高分辨率卫星图像中建筑物提取及变化检测研究
文献类型：学位论文
| 作者 | 段静辉 |
| 学位类别 | 工学博士 |
| 答辩日期 | 2007-05-28 |
| 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 | 卢汉清 ; 普林特 |
| 关键词 | 高分辨率卫星图像 建筑物提取 变化检测 地理信息系统 High Resolution Satellite Image Building Extraction Change Detection Geographic Information System (GIS) |
| 其他题名 | Research on Building Extraction and Change Detection Based on High Resolution Satellite Image |
| 学位专业 | 模式识别与智能系统 |
| 中文摘要 | 近年来，人们能方便地获取到大量高空间、高光谱分辨率和不同传感器类型的遥感图像，同时图像获取时间及间隔所受到的影响逐渐减小。由于这些图像数据能清晰地反映出与人们生活最密切的城区中的景象，因此高分辨率卫星图像在许多应用领域中受到了人们的广泛关注，如地理信息系统（GIS，Geographic Information System）的制作与更新，灾害分析、城区发展规划以及城市地理导航等领域。分辨率的变化使得现有一些针对中低分辨率图像的算法不能直接用于高分辨率图像。因此，对信息丰富的高分辨率图像的处理技术，如提取地面目标、分析地面目标变化，成为目前图像处理和模式识别领域中的一个热点问题，具有很强的挑战性。本文主要针对城区中建筑物的提取和变化检测展开研究工作。研究中使用了“QuickBird”卫星的高分辨率图像数据和GIS数据。本文工作及贡献主要包括下面三个部分： 1)提出了一种基于分割的建筑物提取算法。在分析了图像中建筑物光谱和几何形状上的特点后，提出了一种简单有效的建筑物模型，适用于大多数的建筑物；提出了一种根据分割结果的几何形状特性来确定最优分割阈值的方案，实现了分割算法阈值自动选取。实验证明算法能检测出图像中的大多数建筑物。 2)提出了一种基于贝叶斯网络模型的建筑物提取算法。本章提出了一个结合分割以及区域融合筛选的建筑物提取框架并构建了一个建筑物贝叶斯网络模型，可以利用建筑物在图像中的不同特征计算区域属于建筑物的概率。算法使用分割算法从图像中提取候选的建筑物屋顶区域，然后采用区域融合和筛选的方式从这些结果中提取出最后建筑物区域。实验结果表明该算法能有效地完成对大部分建筑物的提取。 3)提出了一种基于分类的GIS数据和图像之间的变化检测算法。通过分类的算法来完成GIS数据和图像中建筑物变化检测。该算法不需要对图像进行目标提取，只需要从图像提取一些相关特征，从而降低了计算量，这对于大批量数据的处理是十分有利的。另外，通过GIS数据提供的一些先验信息，可以在分类前对参数进行学习，因此该算法能更准确的发现变化。 |
| 英文摘要 | In recent years, plentiful of remote sensing images have been available for earth observation, covering different types of sensors, different spatial resolutions, different spectral resolutions and different temporal resolutions. Due to high resolution, most of man-made objects in urban areas can be seen clearly in image. The high resolution satellite images have drawn increasing attention in different fields, such as GIS(Geographic Information System) data creation and updating, disaster monitoring in city, city plan, navigation in city, etc. Most of the algorithms that applied to moderate and low resolution images can not achieve satisfying results on high resolution images, so technologies adapted to high resolution images with complex contents have been researched and became an active and challenging research field which includes man-made objects detection, change detection and etc. The objective of the present work is to detect buildings in urban areas and analyze their changes between GIS and image. “QuickBird” satellite image and building layer of GIS data are used in our experiments. The main contributions of this dissertation thesis include: 1)Proposed a building extraction method based on segmentation. The main contributions of this method lie in two aspects: First, a building model adapted to describe most of the buildings in images has been introduced; Second, threshold optimization is proposed according to shape of segmented results and threshold can be acquired automatically. Experimental results demonstrated the effectiveness of the proposed method. 2)Proposed a building extraction method based on Bayesian Network. First, candidate building areas are extracted using Grab Cut algorithm, final buildings are acquired from the candidates with high probability. Building Bayesian Network is applied to calculate the probability of buildings. Experimental results demonstrated that the proposed algorithm has an encouraging detection performance. 3)Proposed a change detection method between GIS data and image. In the method, building change between GIS data and image can be detected by using classification algorithm. Compared to many methods that extract the buildings before change detection, our method only needs some features related to building. So computing cost of the method is low. Experimental results indicated a good performance of the method. |
| 语种 | 中文 |
| 其他标识符 | 200118014604865 |
| 源URL | [http://ir.ia.ac.cn/handle/173211/5978]  |
| 专题 | 毕业生_博士学位论文 |
| 推荐引用方式 GB/T 7714 | 段静辉. 高分辨率卫星图像中建筑物提取及变化检测研究[D]. 中国科学院自动化研究所. 中国科学院研究生院. 2007. |
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。