面向对象分类方法在土地利用调查中的应用研究

发布时间：2018-10-19 09:58

【摘要】：传统的土地利用调查方法更新周期长、工作量大、效率低且成本较高,遥感影像分类技术能够快速、准确地分析土地利用情况,掌握真实的土地基础数据,在土地利用调查中具有广泛的应用前景和巨大的应用价值。而传统基于像元的分类方法完全依靠地物的光谱信息,忽略了高分辨率影像中丰富的空间信息导致分类结果会受到异物同谱现象和椒盐噪声的干扰。针对这一问题,本文将面向对象分类方法引入到土地调查中,这种方法很好的克服了传统分类法的弊端,且其分类结果可直接以矢量多边形的形式输出,能够直接导入到GIS环境下进行编辑处理与应用分析。本文以2010年的IKONOS高分辨率遥感影像作为数据源,以东北大学及附近区域作为实验区,针对面向对象分类方法进行了大量的实验,重点从以下几个方面做了分析和研究：1)在分割过程中加入纹理滤波和边缘检测层,并通过多次试验得到最优分割参数。实验对比各类地物在不同尺度下的RMAS值,确定出每一类地物对应的最佳尺度,结合层次间的拓扑关系,最终建立一个由50、70和90三个尺度层组成的网络结构。2)根据国家现行土地分类标准,结合影像目视解译和实地调查确定研究区包含的类别。分析研究描述地物类别的最佳特征或特征组合,建立分类规则树,并将面向对象法分类结果与传统基于像元法分类结果进行对比。在面向对象分类得到的矢量结果图上对应实地选取地物样本,分别对地物样本的长、宽及面积进行量测与计算。3)采用面向对象技术对两期Landsat-7影像进行分类,基于分类结果构建变化检测规则,提取地物类别的变化图斑,并对变化图斑的面积进行统计分析。实验结果表明,面向对象分类方法的总体分类精度为90.68%,比传统基于像元的最大似然法总体精度提高了18.98%,将其应用到土地利用调查中,实现了土地利用分类过程的自动化。但由于面向对象分类法得到的矢量对象边界多呈现锯齿状,图上量测的部分结果并非其直线距离,致使与实地同名地物间存在一定的误差,故需要对其边界进行平滑处理后方能入库。此外,将面向对象分类技术与变化检测相结合的方法,能够快速、准确的检测出土地的变化信息,为及时完善和更新土地利用数据库提供了先进的技术手段。
[Abstract]:Traditional land use survey methods have long updating period, large workload, low efficiency and high cost. Remote sensing image classification technology can analyze the land use situation quickly and accurately, and master the real land basic data. It has wide application prospect and great application value in land use survey. However, the traditional pixel based classification method completely depends on the spectral information of the ground object, and neglects the abundant spatial information in the high-resolution image, which results in the interference of the classification results from the phenomenon of foreign body isospectrum and the salt and pepper noise. In order to solve this problem, this paper introduces the object-oriented classification method into the land survey. This method overcomes the disadvantages of the traditional classification method, and the classification results can be directly output in the form of vector polygons. Can be directly imported into the GIS environment for editing processing and application analysis. In this paper, the IKONOS high-resolution remote sensing image in 2010 is used as the data source, and the Northeast University and its adjacent areas are used as the experimental areas. A large number of experiments have been carried out on the object-oriented classification method. The following aspects are analyzed and studied: 1) texture filtering and edge detection layer are added in the segmentation process and the optimal segmentation parameters are obtained through several experiments. By comparing the RMAS values of all kinds of ground objects at different scales, the optimal scale of each kind of ground objects is determined, and the topological relationship between layers is combined. Finally, a network structure consisting of three scales of 50, 70 and 90 is established. 2) according to the current national land classification standards, combined with image visual interpretation and field investigation, the types of the study area are determined. In this paper, the best feature or combination of features to describe ground objects is analyzed, and the classification rule tree is established, and the results of object-oriented classification are compared with the traditional classification results based on pixel method. In the vector result map of object oriented classification, the object samples are selected in the field, and the length, width and area of the object samples are measured and calculated respectively. 3) the object oriented technology is used to classify the two Landsat-7 images. Based on the classification results, the change detection rules are constructed, and the change patterns of the ground objects are extracted, and the area of the change patterns is analyzed statistically. The experimental results show that the overall classification accuracy of the object-oriented classification method is 90.68, which is 18.98 higher than that of the traditional maximum likelihood method based on pixel. It is applied to the land use survey to realize the automation of the land use classification process. However, the boundary of vector objects obtained by object-oriented classification is serrated, and some of the measured results on the map are not linear distance, which leads to some errors between the vector objects and the objects of the same name in the field. Therefore, it is necessary to smooth the boundary before it can be stored. In addition, the combination of object-oriented classification technology and change detection can quickly and accurately detect the land change information, and provide an advanced technical means for the timely improvement and updating of land use database.
【学位授予单位】：东北大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：P237

【参考文献】