基于异质度的地理国情数据变化检测方法

发布时间：2018-03-06 07:42

本文选题：变化检测　切入点：像斑类别异质度　出处：《成都理工大学》2017年硕士论文　论文类型：学位论文

【摘要】：地理国情数据是各种地理信息的综合和集成,反映国家自然资源、国家生态环境发展现状,成为制定国家和区域发展战略与发展规划的重要基础,能够表达地表特征和地表变化情况,通过一些变化检测的方法可以将两个时期的地理国情数据进行检测并得到检测结果。本文主要研究地理国情数据变化检测的方法,首先介绍了变化检测理论、数据源、基本单元和变化检测一般流程,变化检测的数据源和基本单位在变化检测过程中是非常重要的研究对象。变化检测的一般流程又分为数据预处理、地理特征的提取、变化检测和精度评估分析等。本文变化检测采用的数据源是旧时期的矢量图和新时期的遥感影像,变化检测的基本单元是像斑。变化检测方法是基于异质度的地理国情数据变化检测方法,该方法为了实现矢量图与遥感影像的自动变化检测,提出了一种基于像斑类别异质度的矢量图与遥感影像变化检测方法。以矢量图为约束,对遥感影像进行影像分割获取像斑,采用标记分水岭算法实现矢量图约束的影像分割,在分水岭的约束下利用传统的标记分水岭算法实现影像分割;提取分割后像斑的直方图作为像斑的特征,利用直方图表达像斑的特征后,像斑的特征距离就转换为直方图的距离,再用G统计量的算法计算直方图距离,利用像斑与旧时期同类别像斑特征距离的均值来构建像斑的类别异质度,像斑类别异质度表示像斑与其旧时期所属地物类别间的异质性;采用大津法获取各类别的异质度阈值,将像斑的类别异质度与所属地类的类别异质度阈值进行比较,实现像斑的变化/未变化的判别。采用C++语言,基于GDAL(Geospatial Data Abstraction Library)开源库与ArcEngine平台,建设了地理国情常态化监测数据变化检测系统。系统实现了栅格与矢量数据的加载与叠加显示、带约束影像分割、像斑特征提取、像斑异质度构建、异质度阈值获取等功能。使用2014年的地理国情普查的地表覆盖矢量数据与2016年QucikBird遥感影像做实验分析,为了验证本文方法的有效性,将本文方法同基于像斑灰度均值的方法进行了对比,正确率提升了0.08,误检率与漏检率分别下降了0.27、0.12。
[Abstract]:Geographical national data is the synthesis and integration of all kinds of geographic information, reflecting the development status of national natural resources and national ecological environment, and becoming an important basis for the formulation of national and regional development strategies and development plans. It can express the surface characteristics and the change of the surface, and can detect the data of two periods by some methods of change detection. This paper mainly studies the method of the change detection of the data of the geographical situation. First of all, it introduces the theory of change detection, data source, basic unit and general process of change detection. The data sources and basic units of change detection are very important research objects in the process of change detection. The general process of change detection is divided into data preprocessing, extraction of geographical features, The data sources used in this change detection are vector images of the old period and remote sensing images of the new period. The basic unit of change detection is the image spot. The change detection method is based on the heterogeneity of the geographic national conditions data change detection method, in order to realize the automatic change detection of vector image and remote sensing image, In this paper, a method of vector image and remote sensing image change detection based on the heterogeneity of image spot category is proposed. With vector image as constraint, remote sensing image is segmented to obtain image spot, and image segmentation with vector constraints is realized by using marked watershed algorithm. Under the constraint of watershed, the traditional marking watershed algorithm is used to realize image segmentation, and the histogram of image spot after segmentation is extracted as the feature of image spot, and the feature of image spot is represented by histogram. The feature distance of the image spot is transformed into the distance of the histogram, then the histogram distance is calculated by the algorithm of G statistic, and the class heterogeneity of the image spot is constructed by using the mean value of the feature distance between the image spot and the same image spot of the old period. The heterogeneity degree of image spot category indicates the heterogeneity between image spot and its old feature category, and the threshold value of different heterogeneity degree is obtained by using the method of Dajin, and the class heterogeneity degree of image spot is compared with that of the category heterogeneity of belongs to ground category. Based on the GDAL(Geospatial Data Abstraction Library open source library and ArcEngine platform, the system of detecting the change of the change of the normal monitoring data of the geographical national conditions is built. The system realizes the loading and superposition display of the raster and vector data, which is based on the C language and the open source library and ArcEngine platform of GDAL(Geospatial Data Abstraction Library. The functions of constrained image segmentation, image spot feature extraction, image spot heterogeneity construction, heterogeneity threshold acquisition and so on are analyzed experimentally using the ground cover vector data from the 2014 General Survey of Geographical conditions and the 2016 QucikBird remote sensing image. In order to verify the effectiveness of this method, the method is compared with the method based on the gray mean of image spot. The correct rate is increased by 0.08, and the false detection rate and the missed detection rate are reduced by 0.27 and 0.12 respectively.
【学位授予单位】：成都理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：P208

【参考文献】