中国农村地区建筑物样本及标注无人机影像数据集  

A dataset of building sampling and labeling of UAV images in rural China

作　　者：刘耀辉 杨新月 李嘉禾 程昊 周洁 范熙伟[3,4] 张昊宇 李晓丽[6] 齐文华 李志强[6] 聂高众 徐南 付博[8] 姚国标 于明洋 孟飞 靳奉祥 LIU Yaohui;YANG Xinyue;LI Jiahe;CHENG Hao;ZHOU Jie;FAN Xiwei;ZHANG Haoyu;LI Xiaoli;QI Wenhua;LI Zhiqiang;NIE Gaozhong;XU Nan;FU Bo;YAO Guobiao;YU Mingyang;MENG Fei;JIN Fengxiang(School of Surveying and Geo-Informatics,Shandong Jianzhu University,Jinan 250101,P.R.China;School of Remote Sensing and Information Engineering,Wuhan University,Wuhan 430079,P.R.China;Institute of Geology,China Earthquake Administration,Beijing 100029,P.R.China;Key Laboratory of Seismic and Volcanic Hazards,China Earthquake Administration,Beijing 100029,P.R.China;Institute of Engineering Mechanics,China Earthquake Administration,Harbin 150080,P.R.China;China Earthquake Networks Center,Beijing 100045,P.R.China;College of Marine Science and Engineering,Nanjing Normal University,Nanjing 210023,P.R.China;Institute of Applied Ecology,Chinese Academy of Sciences,Shenyang 110016,P.R.China)

机构地区：[1]山东建筑大学测绘地理信息学院,济南250101 [2]武汉大学遥感信息工程学院,武汉430079 [3]中国地震局地质研究所,北京100029 [4]中国地震局地震与火山灾害重点实验室,北京100029 [5]中国地震局工程力学研究所,哈尔滨150080 [6]中国地震台网中心,北京100045 [7]南京师范大学海洋科学与工程学院,南京210023 [8]中国科学院沈阳应用生态研究所,沈阳110016

出　　处：《中国科学数据（中英文网络版）》2022年第2期179-191,共13页China Scientific Data

基　　金：国家对地观测科学数据中心开放基金项目(NODAOP2020008);山东省自然科学基金项目(ZR2021QD074);河北省地震动力学重点实验室开放基金项目(FZ212203)

摘　　要：农村建筑物是观察农村土地变化和经济发展的基础资料。中国作为农业大国,从高空间分辨率遥感影像上及时、准确提取农村建筑物,对于农村发展至关重要。近年来,随着计算机视觉和运算能力的迅速发展,深度学习以其自动学习特征、适用性强等优点,已在建筑物自动提取等领域取得较好效果。深度学习通常需要大量的训练数据。目前,深度学习提取建筑物常用的数据集以国际上开源建筑物数据集为主,包括Massachusetts、INRIA、WHU等。这些数据集大多基于国外建筑物,缺乏开源、高精度、覆盖范围广、贴切我国农村地区建筑主体结构的建筑物样本数据。为此,本研究基于2017-2020年在陕西渭南、江苏淮安、四川康定、广东汕尾、广东惠州、新疆阿图什、吉林松原等多个中国农村地区采集的无人机航拍图像,制作并开放共享本数据集。本数据集空间分辨率高,基本涵盖我国农村地区房屋建筑的主体结构类型,可应用深度学习方法进行建筑物提取,并可进一步结合具体研究目标进行空间分析和研究,对于国土部门统筹城乡发展和美丽乡村建设具有重要意义和应用价值。Buildings are one of the most important means to observe rural land changes and economic development in rural areas.For an agricultural country like China,timely and accurate extraction of rural buildings from high-resolution remote sensing images is crucial to rural development and planning.Recently,with the great advancements of computer vision and computing capabilities,deep learning has achieved considerable achievements in many applications such as building extraction due to its automatic learning features and strong applicability.Deep learning usually requires large amounts of training data.At present,the datasets commonly used in deep learning to identify buildings are mainly internationally open-source building datasets,including Massachusetts,INRIA,WHU,etc.These datasets are based on foreign buildings,lacking sampling data of buildings that are open-source,high-precision,wide-covered,and not suitable for the architectural style of rural areas in China.Hence,we proposed an open-resource dataset named “A dataset of building sampling and labeling of UAV images in rural China”.This dataset is based on the unmanned aerial images(UAV) collected in Weinan City,Shaanxi Province,Huai’an City,Jiangsu Province,Kangding City,Sichuan Province,Shanwei City,Guangdong Province,Huizhou City,Guangdong Province,Atushi City,Xinjiang,Songyuan City,Jilin Province,and other rural areas in China from 2017 to 2020.This dataset of high spatial resolution can basically represent the building characteristics in rural China.It can be applied to building extraction using deep learning methods as well as be combined with further research for spatial analysis.Furthermore,it is of great significance for the overall planning for the development of urban and rurals and the beautiful countryside construction propose by the State Land Administration of China.

关 键 词：遥感 无人机 中国农村 建筑物 样本及标注 数据集 深度学习 

分 类 号：P237[天文地球—摄影测量与遥感]