MFBFS:A fine-grained building feature set for high-resolution multispectral remote sensing images
Building information extraction from remote sensing images plays an essential role in urban information management and disaster prevention and mitigation.This study establishes a fine-grained building feature set,namely,MFBFS,for high-resolution multispectral remote sensing images.MFBFS uses the domestically produced Gaofen-2 multispectral remote sensing images as data source and selects 21 districts and counties with concentrated buildings in various disaster zones in China,covering 3668 km2 as the study area.These regions include Yongjia County,Xuwen County,and Wanning City in the southeastern coastal disaster belt;Ning'an City,Kaiyuan City,Laiyuan County,Shouguang City,Xinxiang County,Lujiang County,Hengdong County,and Songbei District in the eastern disaster belt;Daning County,Enshi City,Tengchong City,and Shuicheng County in the central disaster belt;Kashgar City,Yizhou District,and Pingluo County in the northwest disaster belt;and Diebu County,Yushu City,and Milin County in the Qinghai-Tibetan disaster belt.To obtain high-quality and high-resolution remote sensing images,a series of preprocessing procedures was applied to the Gaofen-2 images.Initially,poor-quality images were removed,followed by radiometric and orthorectification corrections on multispectral and panchromatic images,respectively.Finally,the panchromatic images were fused to enhance the spatial resolution of the multispectral images,resulting in a spatial resolution of 0.8 m.Seventeen feature components were generated from four perspectives:spectral,texture,edge,and index.Spectral features include features from the blue,green,red,and near-infrared bands.Texture features consist of contrast,dissimilarity,homogeneity,correlation,angular second moment,local binary pattern,and histogram of oriented gradients.Edge features comprise first-order and multi-order edge characteristics.Index features include building,shadow,vegetation,and water indexes.MFBFS encompasses over 260000 building instances,ensuring high intra-class diversity in terms of size,shape,color,orientation,background,and structural type.These instances are classified into three structural types,namely,steel and reinforced concrete,masonry,and block stone structures,significantly reflecting the abilities of buildings to resist disasters and their usable lifespans.The fine-grained design will cause the task of extracting buildings through remote sensing to play a greater role,particularly in pre-disaster loss prediction and post-disaster loss assessment in the disaster field.Rigorous quality control processes and field inspections were conducted to ensure the high accuracy of ground truth values.This procedure involved adherence to interpretation standards and inviting data inspectors and remote sensing image experts to assess the quality of remote sensing images and corresponding ground truth values.Ultimately,191 GB of high-quality feature and label data were obtained.Each of the 17 feature components comprises 11005 512×512-sized feature maps with a spatial resolution of 0.8 m,uniformly expanded to a value range of[0,1].Initial deep learning experiments demonstrate the effectiveness of MFBFS.This feature set,available for download athttps://github.com/WangZhenqing-RS/MFBFS,provides robust data support for fine-grained building structure extraction research and promotes the development of domestic high-resolution remote sensing data applications.
NETL
NSTL
万方数据
