基于无人机影像的农村地区建筑物自动提取方法

doi:10.3969/j.issn.0253-4967.20240171

摘要/Abstract

摘要：

地震灾害损失预评估是做好地震应急准备和降低灾害损失的重要内容, 而房屋建筑高度、面积、分布位置及结构类型数据是预评估工作的基础。为实现房屋建筑物数据的自动提取, 文中重点围绕农村地区房屋情况开展研究, 首先利用无人机获取研究区可见光遥感图像, 经预处理、设定参数、空三加密、拼接等步骤后, 生成研究区的数字表面模型、高程模型及正射影像数据。在房屋数据提取过程中, 从提取效率及精度的角度分别对比了基于高程信息的提取方法与监督分类方法, 其中基于高程提取的方法在差值分析中, 经统计像元均值和标准差, 确定高差范围为2.7~12m, 面积阈值设定为4.8m², 提取面积约为80 566.8m², 经对比分析, 提取精度达90%以上, 相较于监督分类方法在房屋面积、分布位置等方面效果更佳, 整体提取精度受地物类型影响不大。在房屋结构类型占比方面, 利用以上房屋提取结果, 结合实地抽样调查情况, 进一步解译房屋结构类型, 用于预评估结果的产出。文中采用的方法可应用于房屋建筑物数据的提取, 能提高地震灾害损失预评估工作的效率和精度。

关键词: 预评估, 建筑物数据, 可见光遥感, 高程信息, 房屋结构类型

Abstract:

The pre-assessment of earthquake disaster losses is an important component in preparing for earthquake emergencies and reducing disaster losses, and it is also an important basis for rapid assessment after the earthquake. The data of building height, area, distribution position, and structure type are the basis of pre-evaluation work. To enable automatic extraction of building data, this paper focuses on housing in rural areas and adopts a method based on low-altitude UAV photogrammetry to continuously improve the accuracy of building data extraction, thereby providing important basic data support for the pre-assessment of earthquake disaster losses.

Because of the large number of houses in the pre-assessment area, it is impossible to complete the field investigation one by one. First, the visible light remote sensing image of the study area is obtained by UAV, and then the accurate extraction of building data is completed according to the steps of data preprocessing, difference analysis, automatic extraction of building information, and result verification. In data preprocessing, the digital surface model, elevation model and orthophoto data of the study area are generated through data analysis, parameter setting, spatial three-dimensional encryption, stitching and other steps. In the process of building data extraction, the extraction method based on elevation information is compared with the supervised classification method in terms of extraction efficiency and accuracy. The extraction method based on elevation information is based on the idea of making a difference between DSM and DEM. By making mean and standard deviation of pixels in different ranges, the height difference range is determined, and then the misclassified areas are eliminated according to the threshold. The error and noise problems are reprocessed to extract more accurate building information. In order to verify the accuracy of data extraction, in addition to superimposing and comparing the extracted images with DOM images, a field sampling survey was conducted to further verify the extracted results. In the proportion of housing structure types, the housing structure types are further interpreted by using the above housing extraction results and combining them with the field sampling survey, which is used for the output of pre-evaluation results.

In the supervised classification method, building information is mainly obtained by identifying the building features in the DOM images. The results show that this method can extract buildings accurately, and the extracted building area is 94 051.3m². However, there is a wrong extraction for cement roads with similar brightness and features to the roof, and there is a phenomenon of missing extraction for some old roof houses. In addition, the height information of buildings in this method can only be obtained by visual interpretation of inclined images in the same position, and then the number of buildings can be judged according to the interpreted height, and then the building area can be calculated. Based on the method of height difference extraction, in the difference analysis, the average and standard deviation of pixels are counted according to the height difference greater than 2 meters, and the range of height difference is determined to be 2.7 meters to 12 meters. Then, taking the house area as the reference value, the threshold value is set to 4.8m² to eliminate the misclassified area. and the problems of error and noise are reprocessed, and finally, the house area was 80 566.8m². The results show that this method is not only more accurate in extracting information, but also more convenient in obtaining the height, area and distribution position of houses. By comparing and analyzing the remote sensing images obtained by UAV with the results of the field investigation, the accuracy of house extraction is evaluated. The results show that the extraction accuracy of the method based on elevation information on the ground buildings is over 90%. Compared with the supervised classification method, there are relatively few misclassified areas, which can identify the building information more accurately. The extracted house information can be used to further interpret the proportion of houses with different structural types in the study area and improve the efficiency of interpretation.

In this study, rural areas are selected as the research area. To obtain accurate building data information in the research area, orthographic and oblique remote sensing images of the research area are collected by unmanned aerial vehicles, and two automatic extraction methods are compared and analyzed, which verifies the accuracy of the extraction methods. Compared with the supervised classification method, the extraction method based on elevation information adopted in this paper has higher reliability, higher extraction accuracy, and better extraction effect when the ground object type is complex. The overall extraction accuracy is not affected by the ground object type, and the building information in the research area can be extracted quickly and accurately. The extraction results can be used to interpret the area and proportion of buildings with different structural types, and can supplement and improve the spatial distribution database of buildings in the study area. This method realizes the automatic extraction of building information in the study area, improves the efficiency and accuracy of the pre-assessment work, and has been applied in the pre-assessment of earthquake disaster losses in rural areas of several districts and counties.

Key words: pre-assessment, building data, visible light remote sensing, elevation information, type of building structure

李晓阳, 韩贞辉, 李晓慧, 谢恒义. 基于无人机影像的农村地区建筑物自动提取方法[J]. 地震地质, 2026, 48(3): 892-906.

LI Xiao-yang, HAN Zhen-hui, LI Xiao-hui, XIE Heng-yi. RESEARCH ON AUTOMATIC EXTRACTION METHOD OF BUILDINGS IN RURAL AREAS BASED ON UAV IMAGES[J]. SEISMOLOGY AND GEOLOGY, 2026, 48(3): 892-906.

图/表 11

图1 建筑物自动提取技术流程图

Fig. 1 Flow chart of automatic building extraction technology.

图2 遥感图像处理流程图

Fig. 2 Remote sensing image processing flow chart.

图3 拼接后的研究区DOM数据

Fig. 3 DOM data of the research area after splicing.

图4 面向对象的遥感影像分类框架

Fig. 4 Object-oriented classification framework for remote sensing images.

图5 差值分析流程图

Fig. 5 Difference analysis flow chart.

图6 基于面向对象的监督分类结果图

Fig. 6 Object-oriented supervised classification result diagram.

图7 经目视解译修正后建筑物提取结果图

Fig. 7 The result map of building extraction after visual interpretation and correction.

图8 DSM与DEM差值处理结果图

Fig. 8 DSM and DEM difference processing result chart.

图9 基于高差提取的研究区建筑物分布图

Fig. 9 Distribution map of buildings in the study area based on height difference extraction.

图10 阈值消除后的建筑物分布图

Fig. 10 A building a distribution map after threshold elimination.

图11 基于高差提取的结果影像与DOM影像叠加对比图

Fig. 11 Superimposed contrast diagram of the result image and the DOM image based on height difference extraction.

参考文献 13

[1]	陈桥驿, 闫宇飞, 黄永芳. 2023. 融合面向对象和深度置信网络的农村建筑物信息提取[J]. 测绘通报, (6): 134-137. DOI
	CHEN Qiao-yi, YAN Yu-fei, HUANG Yong-fang. 2023. Rural building information extraction using object-oriented and deep belief network[J]. Bulletin of Surveying and Mapping, (6): 134-137 (in Chinese). DOI
[2]	程紫燕, 杨斌, 郑树平, 等. 2022. 房屋的遥感影像识别在地震灾害风险预评估中的应用探讨[J]. 山西地震, (2): 1-4, 9.
	CHENG Zi-yan, YANG Bin, ZHENG Shu-ping, et al. 2022. Application of house remote sensing image recognition in earthquake disaster risk pre-assessment[J]. Earthquake Research in Shanxi, (2): 1-4, 9 (in Chinese).
[3]	杜浩国, 张方浩, 卢永坤, 等. 2021. 基于无人机影像的面向对象建筑物结构分级提取方法研究[J]. 地震研究, 44(2): 262-274.
	DU Hao-guo, ZHANG Fang-hao, LU Yong-kun, et al. 2021. Research on object-oriented building structure classification extraction method based on UAV image[J]. Journal of Seismological Research, 44(2): 262-274 (in Chinese).
[4]	范熙伟, 聂高众, 邓砚, 等. 2017. 基于摄影测量技术的房屋提取方法: 以中国西部地区乡村为例[J]. 地震地质, 39(4): 805-818. doi: 10.3969/j.issn.0253-4967.2017.04.014.
	FAN Xi-wei, NIE Gao-zhong, DENG Yan, et al. 2017. The extraction of hose distribution based on photogrammetry method: Taking the countryside in the west of China, for example[J]. Seismology and Geology, 39(4): 805-818 (in Chinese). DOI
[5]	李东臣, 任俊杰, 张志文, 等. 2022. 基于高分辨率无人机影像的地震地表破裂半自动提取方法: 以2021年 M_S7.4 青海玛多地震为例[J]. 地震地质, 44(6): 1484-1502. doi: 10.3969/j.issn.0253-4967.2022.06.008.
	LI Dong-chen, REN Jun-jie, ZHANG Zhi-wen, et al. 2022. Research on semi-automatic extraction method of seismic surface ruptures based on high-resolution UAV image: Taking the 2021 M_S7.4 Maduo earthquake in Qinghai Province as an example[J]. Seismology and Geology, 44(6): 1484-1502 (in Chinese). DOI
[6]	刘超, 杜鹏, 王银, 等. 2021. 基于无人机摄影测量技术研究有无地面控制点的差异性在地震方面的应用[J]. 华南地震, 41(1): 84-93.
	LIU Chao, DU Peng, WANG Yin, et al. 2021. Research on the application of the difference with or without ground control points in earthquake based on UAV photogrammetry technology[J]. South China Journal of Seismology, 41(1): 84-93 (in Chinese).
[7]	路尧, 丁小牛. 2021. 无人机遥感在地震灾害损失预评估中的应用[J]. 测绘通报, (S1): 170-172.
	LU Yao, DING Xiao-niu. 2021. Application of UAV remote sensing in earthquake disaster loss pre-assessment[J]. Bulletin of Surveying and Mapping, (S1): 170-172 (in Chinese).
[8]	宋平, 聂高众, 邓砚, 等. 2016. 基于现场调查的地震灾害损失预评估: 以云南省德宏傣族景颇族自治州为例[J]. 地震地质, 38(4): 1148-1159. doi: 10.3969/j.issn.0253-4967.2016.04.027.
	SONG Ping, NIE Gao-zhong, DENG Yan, et al. 2016. Pre-assessment of earthquake-induced losses based on field survey and kilometer grid dataset: A case study from Dehong dai-jingpo autonomous prefecture[J]. Seismology and Geology, 38(4): 1148-1159 (in Chinese). DOI
[9]	孙稳, 何宏林, 魏占玉, 等. 2019. 基于无人机航测获取高分辨率DEM数据的断层几何结构精细解译与分析: 以海原断裂唐家坡为例[J]. 地震地质, 41(6): 1350-1365. doi: 10.3969/j.issn.0253-4967.2019.06.003.
	SUN Wen, HE Hong-lin, WEI Zhan-yu, et al. 2019. Interpretation and analysis of the fine fault geo-metry based on high-resolution DEM data derived from UAV photogrammetric technique: A case study of Tangjiapo site on the Haiyuan Fault[J]. Seismology and Geology, 41(6): 1350-1365 (in Chinese). DOI
[10]	吴浩霖, 聂高众, 范熙伟, 等. 2019. 基于小型无人机航拍影像的地表房屋建筑提取算法对比: 以新疆阿图什市琼哈拉峻村为例[J]. 地震研究, 42(2): 236-244, 306.
	WU Hao-lin, NIE Gao-zhong, FAN Xi-wei, et al. 2019. Comparison of algorithms for surface buildings extraction based on UAV images: Taking Qiongharajun village in Atushi city, Xinjiang as an example[J]. Journal of Seismological Research, 42(2): 236-244, 306 (in Chinese).
[11]	于浩, 贾晓东, 庞巧, 等. 2023. 基于无人机影像的辽宁农居建筑物提取技术与应用研究[J]. 震灾防御技术, 18(1): 96-106.
	YU Hao, JIA Xiao-dong, PANG Qiao, et al. 2023. Research on extraction technology and application of rural residential buildings in Liaoning Province based on UAV image[J]. Technology for Earthquake Disaster Prevention, 18(1): 96-106 (in Chinese).
[12]	张雪华, 王晓青, 杜晓霞, 等. 2019. 基于无人机遥感影像及其点云特征的建筑物震害提取[J]. 地震研究, 42(2): 230-235, 305.
	ZHANG Xue-hua, WANG Xiao-qing, DU Xiao-xia, et al. 2019. Extraction of buildings’ seismic damage based on remote sensing images of UAV and its point clouds characteristic[J]. Journal of Seismological Research, 42(2): 230-235, 305 (in Chinese).
[13]	朱达邈, 王东明. 2021. 基于遥感影像的建筑物空间分布方法在地震灾害损失预评估中的应用[J]. 震灾防御技术, 16(3): 510-518.
	ZHU Da-miao, WANG Dong-ming. 2021. Application of building spatial distribution method based on remote sensing image in earthquake disaster loss pre-assessment[J]. Technology for Earthquake Disaster Prevention, 16(3): 510-518 (in Chinese).