论文部分内容阅读
目的:资源一号02C星搭载国产遥感卫星序列中为数不多的高性能传感器之一,获取大量的影像数据。然而,在空间分辨率相对较高,光谱分辨率比较低的情况下,城市土地覆盖分类势必存在一定问题。如何深度挖掘影像光谱和空间信息,建立可行的技术方法流程,实现准确的城市土地覆盖分类,进而为其推广应用奠定基础十分必要。创新点:提出光谱与空间领域信息、判别分析、面向对象法结合的技术流程体系(图2),实现城市土地覆盖的准确分类。对分类结果采用基于点和图斑面积的两种验证方法进行验证。方法:计算图像纹理、空间自相关特征、形状指数、植被指数、不透水面含量等信息,与光谱信息结合,经过判别分析和相关分析的筛选,实现面向对象的分类和两种指标的精度评价。结论:根据本文提出的技术路线,可以实现相对准确的城市土地覆盖分类。总体点位精度在92%以上(表2),面积精度达到82%以上,误差通常源自住宅和裸土的混淆。影像数据在城市土地覆盖分类方面非常有效。
OBJECTIVE: Resource One 02C is equipped with one of the few high performance sensors in domestic remote sensing satellite series to acquire a large amount of image data. However, in the case of relatively high spatial resolution and relatively low spectral resolution, there is bound to be some problems in the classification of urban land cover. How to mine the spectral and spatial information in depth, establish feasible technical methods and processes, and achieve accurate classification of urban land cover so as to lay the foundation for its popularization and application. Innovative point: Put forward the technical flow system of spectrum and space domain information, discriminant analysis and object-oriented method (Figure 2), and realize the accurate classification of urban land cover. The classification results were validated using two validation methods based on the area of points and patch areas. Methods: The information of image texture, spatial autocorrelation, shape index, vegetation index and impervious surface were calculated and combined with the spectral information. After discriminant analysis and correlation analysis, object-oriented classification and accuracy evaluation of two indexes were realized . Conclusion: According to the technical route proposed in this paper, we can achieve a relatively accurate classification of urban land cover. Overall accuracy of more than 92% points (Table 2), the accuracy of more than 82% area, the error usually comes from the confusion between residential and bare soil. Image data is very effective in classifying urban land cover.