基于多尺度影像分割的面向对象城市土地覆被分类研究——以马来西亚吉隆坡市城市中心区为例

Object-oriented Urban Land-cover Classification of Multi-scale Image Segmentation Method——a Case Study in Kuala Lumpur City Center,Malaysia

城市受人类活动影响比较大,结构组成比较复杂,对该区域进行分类研究存在一些问题。甚高分辨率遥感影像,以其丰富的细节信息为城市土地覆被分类研究提供了可能。本文结合使用甚高分辨率Qu ickB ird遥感影像和激光扫描LIDAR数据,论述了利用多尺度、多变量影像分割的面向对象的分类技术对马来西亚基隆坡市城市中心区的土地覆被分类研究。针对特定地物选择合适的影像分割特征和分割尺度、按照合理的提取顺序逐步进行城市土地覆被信息提取。在建筑物的提取过程中构建了归一化数字表面模型nDSM,使用成员函数将建筑物信息提取出来。精度评价结果表明,利用该方法得到了理想的城市土地覆被分类结果,其分类总精度从常规面向对象分类方法的83.04%上升到88.52%,其中建筑物生产精度从60.27%增加到93.91%。

Urban land cover classification is not easy for its inner complexity resulting from human＇ s impact. The emergency of very high resolution remote sensing images makes it possible to classify the urban land cover for its ample details. The paper is about urban land cover classification method using multi-scale and multi- variable image segmentation method. The innovation of this method lies in selection of proper scale parameter resulting from proper image data and certain classification order. Normalized Digital Surface Model（nDSM） is constructed in the process of building extraction; the threshold of elevation difference is used to extract building. Accuracy assessment result indicates that the ideal urban land cover classification results have been obtained using this method. The total accuracy has increased from 83.04% to 88.52% compared with traditional object oriented classification result. The product accuracy of building increases from 60.27% to 93.91% , which is the highest among all the six objects.