摘要
建筑物轮廓信息是国家基础地理信息的重要组成部分。高精度建筑物轮廓自动提取往往依赖于大规模的建筑物标注样本。为了丰富中国地区建筑物提取数据集,本文研究以高分辨率遥感影像为数据源,采用人工标注与交互式标注相结合的方式构建形成中国典型城市建筑物实例数据集。本数据集包含7260个影像区域样本,共63886栋建筑物,分布在北京、上海、深圳及武汉4个城市。数据集由MS COCO 2017格式的标注文件及相应的建筑物掩膜二值图构成,可为研究高分辨率遥感影像的建筑物检测和提取提供基础数据。
Building contour information is an important part of the national basic geographic information.The performance of building automatic extraction is usually driven by a large number of training samples.To enrich the building extraction datasets of cities in China,we compiled a building instance dataset sourced from high-resolution remote sensing images through the combination of manual annotation and interactive annotation.This dataset consists of the samples in 7,260 regions,with 63,886 building instances in four China’s cities:Beijing,Shanghai,Shenzhen and Wuhan.The annotations of the dataset consist of MS COCO 2017 format files and the corresponding building mask binary maps.This dataset provides fundamental data for the research on building detection of and extraction from high-resolution remote sensing images.
作者
吴开顺
郑道远
陈妍伶
曾林芸
张嘉辉
柴生华
徐文杰
杨永亮
李圣文
刘袁缘
方芳
WU Kaishun;ZHENG Daoyuan;CHEN Yanling;ZENG Linyun;ZHANG Jiahui;CHAI Shenghua;WU Wenjie;YANG Yongliang;LI Shengwen;LIU YuanYuan;FANG Fang(National Engineering Research Center of Geographic Information System,China University of Geosciences,Wuhan 430074,P.R.China;School of Geographic and Information Engineering,China University of Geosciences,Wuhan 430074,P.R.China)
基金
国家对地观测科学数据中心开放基金项目(NODAOP2020015)。
关键词
建筑物提取
实例分割
高分辨率遥感影像
building extraction
instance segmentation
high-resolution remote sensing images