基于面向对象方法的地表覆被分类研究

发布时间：2018-03-20 11:51

本文选题：面向对象　切入点：地表覆被　出处：《成都理工大学》2013年硕士论文　论文类型：学位论文

【摘要】：地表覆被是陆地生命支撑系统的重要组成部分，及时准确地了解地表覆被状况具有重要的意义。遥感技术作为一种综合性的探测手段，在地表覆被分类领域发挥着重要的作用。随着遥感技术的快速发展，高空间分辨率成为一大发展趋势。GeoEye、IKONOS、QuickBird等高分辨率影像在测绘、林业、国土等诸多方面得到了广泛应用。高分辨率遥感影像具有丰富的空间信息和纹理信息，而光谱信息相对较弱，传统的基于像元的影像分类方法往往难以取得满意的效果，，面向对象的分类方法应运而生。面向对象分类方法以对象为最小分类单元，不仅将光谱信息作为分类依据，还充分利用了空间结构和纹理信息，大幅提高了高分辨率遥感影像地表覆被分类的精度与效率。本文在研读大量国内外文献的基础上，论述了面向对象分类方法的原理，分析了面向对象方法较传统分类方法的优势，以eCognition软件为平台，利用研究区QuickBird高分辨率遥感影像进行了地表覆被分类的研究。本文的研究内容和研究成果如下： (1)对研究区各类数据进行了预处理。对地形数据进行处理并生成DEM；对遥感影像数据进行了辐射校正、正射校正、波段优选和融合处理。利用相关性矩阵确定了波段组合，通过比较选择了效果最佳的融合方法。 (2)对最优分割尺度问题进行了详细的探讨，通过对研究区影像进行实验验证，选择了四个分割尺度构建分类层次体系。 (3)充分利用影像对象的光谱、形状、纹理特征等，选取地物的典型分类特征建立合适的分类规则，使用模糊分类方法实现了研究区地表覆被分类。 (4)对研究区面向对象分类结果进行了分类精度评价，并与基于像元的最大似然分类方法进行了分类精度比较。通过比较发现，面向对象的地表覆被分类方法精度更高，同时有效避免了椒盐现象，视觉效果更佳。
[Abstract]:Surface cover is an important part of terrestrial life support system. It is of great significance to understand the land cover situation accurately and timely. Remote sensing technology as a comprehensive means of detection, With the rapid development of remote sensing technology, high spatial resolution has become a major development trend. High resolution remote sensing images have abundant spatial information and texture information, but spectral information is relatively weak. Traditional image classification methods based on pixel are often difficult to achieve satisfactory results. The object oriented classification method takes object as the minimum classification unit, which not only takes spectral information as the basis of classification, but also makes full use of spatial structure and texture information. The accuracy and efficiency of land cover classification in high resolution remote sensing images are greatly improved. On the basis of studying a large number of literature at home and abroad, this paper discusses the principle of object oriented classification method, analyzes the advantages of object oriented method compared with traditional classification method, and takes eCognition software as the platform. Land cover classification is studied by using QuickBird high-resolution remote sensing images in the study area. The research contents and results are as follows:. (1) preprocessing all kinds of data in the study area, processing topographic data and generating demm; performing radiometric correction, orthophoto correction, band optimal selection and fusion processing for remote sensing image data; determining the band combination by using the correlation matrix. By comparison, the best fusion method is selected. In this paper, the optimal segmentation scale is discussed in detail. Four segmentation scales are selected to construct the classification hierarchy through the experimental verification of the image in the study area. 3) make full use of the spectral, shape and texture features of the image object, select the typical classification features of the ground objects to establish the appropriate classification rules, and use the fuzzy classification method to realize the land cover classification in the study area. Finally, the classification accuracy of object oriented classification is evaluated and compared with the maximum likelihood classification method based on pixel. It is found that the precision of object oriented land cover classification method is higher than that of pixel based maximum likelihood classification method. At the same time effectively avoid salt and pepper phenomenon, visual effect is better.
【学位授予单位】：成都理工大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：P237

【参考文献】