doi:10.16186/j.cnki.1673-9787.2020.2.8

Received:2019/06/08

Revised:2019/07/22

Published:2020/03/15

Research on object-oriented building information extraction method

HE Xiaolu1,2,3,4, LIU Zhenhua1,2,3,4, HU Yueming1,2,3,4

1.The College of Natural Resources and Environment， South China Agricultural University， Guangzhou  510642 ， Guangdong， China;2.Key Laboratory for Redevelopment of Construction Land， Ministry of Land and Resources， Guangzhou  510642 ， Guangdong， China;3.Guangdong Provincial Key Laboratory of Land Use and Consolidation， Guangzhou  510642 ， Guangdong， China;4.Guangdong Province Engineering Research Center for Land Information Technology， Guangzhou  510642 ， Guangdong， China;5.College of Agriculture and Animal Husbandry， Qinghai University， Xining  810016 ， Qinghai， China;6.School of Resources and Environment， University of Electronic Science and Technology of China ， Chengdu  610054 ， Sichuan， China

Abstract:It is important to identify buildings effectively to better carry out urban construction and planning. To solve the problem that current research techniques are difficult to achieve high-precision extraction of build- ings ， a method based on the combination of object-oriented and RedEdge band rule and sample-based object-oriented was proposed to extract urban building information. Firstly， the fusion data of panchromatic and multi- spectral of worldview 2 image was adopted to segment the scale， a double-layer rule knowledge base was built to extract the building information according to the spectral characteristics and shape characteristics of buildings ，digital surface model （ DSM） and texture features of RedEdge band of worldview 2 image. Secondly，the sample-based object-oriented method was used to extract building information from the worldview2 image. Finally ，the results of building information obtained by the two object-oriented methods were fused to perform high-precision extraction of buildings. Taking Tianhe district of Guangzhou as an example， the results showed that the classification accuracy of sample-based object-oriented method， rule-based object-oriented method， object-oriented method based on RedEdge band rule and the proposed method were 81.27% ，83. 75% ，87.06% and 91.43 % ， respectively. The results showed the method of combining object-oriented classification based on RedEdge band rule and sample-oriented classification was more accuracy than the other three methods，which provided an effective means for building information recognition of high-resolution remote sensing image.

Key words:building information extraction;object-oriented;Red Edge band;worldview 2

基于面向对象的建筑物信息提取方法研究_贺晓璐.pdf