面向对象的高分辨率遥感影像数据的道路提取流程——地理国情普查数据的自动提取

Objects-oriented Method of Data Extraction in National Geographic Census:A Case Study of Road

为了提高地理国情普查数据生产的自动化程度,针对高分辨率遥感影像的特征,以道路为例,基于面向对象的思想构建了一套通用性较高的数据提取方法流程,并以北京门头沟、河南郑州市作为实验区,证实自动提取规则的有效性。研究结果表明,郊区道路的自动化提取精度比城区道路更高;大比例尺基础数据的引入,可屏蔽掉缓冲区外非道路要素的干扰,提高分类精度。

Taking the case of road,a general data extraction workflow of land cover data is proposed to reduce manual labor and subjectivity in the national geographic census.Taking advantage of recent advances in object-oriented image analysis,the optimal rule set of object metrics used to extract road is also constructed based on the very high resolution remote sensing imagery.Experimental results indicate that the extraction model achieved higher accuracy of roads in the suburb,compared to the one in the urban area.With the combination of large-scale geographic data,the interference by spectral similarities of road with other landscape elements outside of road buffer zone can be eliminated,which is helpful to significantly enhance the classification accuracy.