研究目的：依据《土地利用现状分类》，利用QuickBird高空间分辨率遥感影像，采用面向对象的影像分析方法对黄土高原实验区土地利用进行分类，探索提高土地资源数量及质量信息挖掘的广度、深度、自动化程度与准确度，提高工作效率，为国土资源行业的土地资源遥感监测工作提供技术参考。研究方法：在光谱和纹理分类的基础上，提出利用坡度空间谱和对象空间关系构建决策规则的方法；通过对DEM数据的学习，可以得到对象的高程、坡度和坡度空间谱；通过空间方位、距离和拓扑关系，可以分析对象之间的位置关系；最后将学习所得的规则加入到土地利用分类的规则集中。研究结果：相对于完全基于光谱和纹理信息的分类方法，本方法使分类总精度和KAPPA系数分别从76％和0．71提高到87％和0．86。研究结论：坡度空间谱与空间关系规则的应用能有效减少混分，提高分类精度，对于黄土高原区域土地利用分类规则集的构建具有参考价值。%The purpose of this study is to classify the land use through the object-oriented image analysis method with the QuickBird images. It focuses on the scope, depth, automation degree, and accuracy in order to provide the technological reference for land monitoring through remote sensing. The method used is the newly developed decision-rule method based on the slope spatial spectrum and spatial relations. The slope spatial spectrum and the spatial distribution can be obtained through DEM data. The distance, direction and topological spatial relations can beused to analyze the spatial relation between objectives. The results showed that the slope spatial spectral and spatial relationship can effectively reduce the mixing classifications and improve the classification accuracy. The slope spatial spectrum can reduce misclassification between dry land and grass land. The paper concludes that the spatial relationship can correctly extracted rural homestead which have been error assigned to dry land, and irrigated land,forest land, mining land and inland beaches in the wrong classification has also been resolved. The results for the Loess Plateau of land use classification rules set built have some reference value.
展开▼
