地球信息科学学报

地球信息科学学报 ›› 2016, Vol. 18 ›› Issue (3): 423-432.doi: 10.3724/SP.J.1047.2016.00423

• 遥感科学与应用技术 • 上一篇    下一篇

基于图分割的高分辨率遥感影像建筑物变化检测研究

施文灶1,2,3,4(), 毛政元1,3,4   

  1. 1. 福州大学 空间数据挖掘与信息共享教育部重点实验室,福州 350002
    2. 福建师范大学光电与信息工程学院,福州 350108
    3. 福州大学地理空间信息技术国家地方联合工程技术研究中心,福州 350002
    4. 福州大学福建省空间信息工程研究中心,福州 350002
  • 收稿日期:2015-05-27 修回日期:2015-09-29 出版日期:2016-03-10 发布日期:2016-03-10
  • 作者简介:

    作者简介:施文灶(1982-),男,博士生,讲师,研究方向为高空间分辨率遥感影像信息提取.E-mail:swz@fjnu.edu.cn

  • 基金资助:
    国家自然科学基金项目(41201427);"十二五"国家科技支撑计划项目(2013BAC08B02-01);福建省教育厅项目(JB14038)

The Research on Building Change Detection from High Resolution Remotely Sensed Imagery Based on Graph-cut Segmentation

SHI Wenzao1,2,3,4,*(), MAO Zhengyuan1,3,4   

  1. 1. Key Lab of Spatial Data Mining & Information Sharing of Ministry of Education, Fuzhou University, Fuzhou 350002, China
    2. College of Photonic and Electronic Engineering, Fujian Normal University, Fuzhou 350108, China
    3. National Engineering Research Centre of Geospatial Information Technology, Fuzhou University, Fuzhou 350002, China
    4. Spatial Information Engineering Research Centre of Fujian Province, Fuzhou University, Fuzhou 350002, China
  • Received:2015-05-27 Revised:2015-09-29 Online:2016-03-10 Published:2016-03-10
  • Contact: SHI Wenzao E-mail:swz@fjnu.edu.cn

RichHTML

PDF (PC)

赞

被引次数

6 | 5

摘要：

建筑物是城市地理数据库中最容易发生变化和最需要更新的部分,其更新工作量巨大,因此开展对高分辨率遥感影像中的建筑物进行自动提取和变化检测研究具有重要的意义.本文以精确提取变化建筑物的位置和轮廓为目标,基于图分割提出一种高分辨率遥感影像建筑物变化检测方法.首先,将遥感影像中的每个像元映射成图的顶点,利用像元之间的距离阈值构造图的边,综合利用位置,灰度和边缘3种特征计算边的权值,将遥感影像的分割转化为图的分割,并用归一化图分割方法得到分割对象集合;然后,以长宽比和矩形度作为约束条件,对2期遥感影像中的分割对象集合进行筛选,提取建筑物对象;最后,根据2期影像中建筑物之间的空间,面积和格局关系识别建筑物的变化类型(包括新增,消失和改建),并对其进行可视化表达.为了验证本文方法的有效性,分别以深圳市的WorldView影像和北京市的QuickBird全色影像为数据源,从中选取13组具有代表性的子影像进行实验.结果表明,本文提出的方法对配准精度较低的影像组具有一定的适应性,容许的配准误差达到20个像元(10 m),平均查准率和平均查全率分别达到93.16%和87.90%.

关键词: 高分辨率遥感影像, 图分割, 建筑物, 变化检测

Abstract:

Buildings are most prone to change and the most in need of update in the urban geographic database, but its workload is heavy. Therefore, it is of great significance to study the automatic extraction and change detection of buildings in high-resolution remotely sensed imagery. In order to extract the precise locations and boundaries of the changed buildings, this article proposed an approach for building change detection from the high-resolution remotely sensed imagery based on the graph-cut segmentation. Firstly, each pixel in the remotely sensed imagery was mapped into the vertex of a graph. The edges of the graph were constructed using a distance threshold between pixels, and their weight values were set according to the comprehensive characteristic calculation using the position, intensity and contour. The segmentation of remotely sensed imagery was transformed into graph partitioning, and the segmentation objects set was obtained through normalized graph-cut segmentation method. Then, buildings were extracted by selecting the segmentation objects sets of two temporal remotely sensed imageries with constraint integrating aspect ratio and rectangularity. Last, the change types of buildings were identified, including the newly added buildings, building disappearance and building reconstruction, and their visualizations were expressed, according to the relationships of spaces, areas and configurations between buildings from the two temporal remotely sensed imageries. For verifying the validity of the proposed method, 13 representative sub-image pairs were chosen from the WorldView images which cover the Shenzhen city and the QuickBird panchromatic images which cover the Beijing city, respectively. Experimental results show that the proposed method has revealed certain adaptability to the image pairs with low registration accuracy, which allows a permissible registration error up to 20 pixels (10 meters), and the average precision rate and recall rate of the proposed method are 93.16% and 87.90% respectively.

Key words: high resolution remote sensing image, graph-cut segmentation, buildings, change detection