基于多源数据与随机森林方法的城市建成区提取——以郑州市为例

Extraction of urban built-up area based on multi-source data and random forest method:take Zhengzhou as an example

基于夜间灯光数据的阈值分割法在城镇建成区提取研究中被广泛应用,但由于夜间灯光数据分辨率低、灯光溢出和阈值分割法无法顾及区域差异等问题,一定程度上影响了该方法的提取精度。以郑州市为例,以LJ1-01与NPP/VIIRS两种夜间灯光影像为主要数据源,结合Landsat8中分辨率遥感影像、网络城市兴趣点(POI)及路网数据,利用随机森林分类方法对郑州市2018年建成区进行提取,参考土地利用数据,对RF分类法与NTL、VANUI、BANUI、PANUI、RANUI指数等阈值法进行对比实验和精度评价,评估基于多源数据的随机森林分类方法在城市建成区提取中的优势。实验表明,RF比阈值法提取的建成区更接近真实建成区且提取精度更高,具有更好适用性;LJ1-01数据提取的效果和精度总体优于NPP/VIIRS数据;在采用RF分类时,各类特征的重要性在不同夜光数据源中表现差异较大。

Threshold segmentation method based on night lighting data is widely used in the research of urban built-up area extraction.However,due to the low resolution of night lighting data,light overflow and the fact that the threshold segmentation method cannot take into account regional differences,the extraction accuracy of this method is affected to some extent.In this paper,taking Zhengzhou as an example,taking LJ1-01 and NPP/VIIRS as the main data sources,combined with Landsat8 medium-resolution remote sensing images,interest point of network city(POI)and road network data,the built-up area of Zhengzhou in 2018 was extracted by random forest(RF)classification method.Referring to land use data,the paper carries out the comparison experiment and accuracy evaluation of RF classification method and threshold method based on NTL,VANUI,BANUI,PANUI and RANUI index,and evaluates the advantages of random forest classification method based on multi-source data in the extraction of urban built-up areas.The experiments show that the built-up area extracted by RF is closer to the real built-up area than the threshold method,with higher extraction accuracy and better applicability.The extraction effect and accuracy of LJ1-01 data are better than that of NPP/VIIRS data.When RF classification is adopted,the importance of various features is quite different in different night lighting data sources.