多尺度分割和深度学习相结合的倾斜摄影三维影像建筑物震害信息提取

帅向华; 荆帅军; 郑向向; 刘钦

doi:10.11939/jass.20220105

多尺度分割和深度学习相结合的倾斜摄影三维影像建筑物震害信息提取

doi: 10.11939/jass.20220105

1.
中国北京 100045 中国地震台网中心
2.
中国郑州 450016 河南省地震局
3.
中国北京 100083 中国自然资源航空物探遥感中心

基金项目: 国家重点研发计划（2018YFC1504500）资助

详细信息

通讯作者:
帅向华，硕士，研究员，主要从事灾害风险评估、地震灾害应急与GIS遥感应用方面的研究，e-mail：shuaixhua@sina.com

中图分类号: P315.9
计量
- 文章访问数: 247
- HTML全文浏览量: 140
- PDF下载量: 113
- 被引次数: 0
出版历程
- 收稿日期: 2022-06-17
- 修回日期: 2022-08-08
- 网络出版日期: 2022-09-09
- 刊出日期: 2022-09-15

Extraction of seismic damage information of buildings from three-dimensional images of oblique photography based on multi-scale segmentation and deep learning

1.
China Earthquake Networks Center，Beijing 100045，China
2.
Henan Earthquake Agency，Zhengzhou 450016，China
3.
China Aero Geophysical Survey & Remote Sensing Center for Natural Resources，Beijing 100083，China

摘要

摘要: 采用多尺度分割和深度学习相结合的方法对震后倾斜摄影三维影像建筑物震害信息进行提取，获取建筑物的屋顶和墙体多种破坏信息。以2017年九寨沟M_S7.0地震后倾斜摄影三维影像为例，依据三维影像建筑物顶面和墙体等进行样本的多尺度分割，样本分为完好建筑物面、破坏建筑物面、其它地物和背景等三类，选取211个100×100像素的样本集对卷积神经网络模型进行训练，采用训练后的模型提取灾区千古情风景区和漳扎镇小学的建筑物震害信息，并将提取结果与目视解译结果进行精度对比，结果显示：破坏建筑物面提取精度分别为65.5%和71.1%，总体分类精度分别为82.1%和84.1%，卡帕（Kappa）系数分别为68.7%和64.9%，表明该方法在倾斜摄影三维影像建筑物震害提取方面具有一定的优势。
- 建筑物震害提取 /
- 多尺度分割 /
- 深度学习 /
- 倾斜摄影
Abstract: The method of combining multi-scale segmentation and deep learning is used to extract the seismic damage information of buildings from three-dimensional images of oblique photography after an earthquake. In this way, the comprehensive damage information of the roof and wall of the building is obtained. Taking the 2017 M_S7.0 Jiuzhaigou earthquake as an example, firstly, based on the three-dimensional image of the roof and wall of the building, the multi-scale segmentation of samples are divided into three categories: intact building surface, damaged building surface, other ground objects and background. Secondly, 211 sample set with 100×100 pixel are selected to train the neural convolutional network model. The trained model is used to extract the seismic damage information of buildings in Qianguqing scenic spot and Zhangzha primary school. Finally, the accuracy of the extracted results is compared with the visual interpretation results. The results show that the extraction accuracy of damaged building surface is 65.5% and 71.1% respectively, the overall classification accuracy is 82.1% and 84.1% respectively, and the Kappa coefficient is 68.7% and 64.9% respectively. The result indicates that this method has certain advantages in seismic damage extraction of buildings from three-dimensional images of oblique photography.
- seismic damage extraction of buildings /
- multi-scale segmentation /
- deep learning /
- oblique photography

HTML全文

图 1 VGGNet-16网络模型结构，模型下方数字为图像大小（宽×高×深）

Figure 1. VGGNet-16 structural framework where the number under the structural is the image size (width×high×deep)

下载: 全尺寸图片幻灯片

图 2 千古情风景区的三维模型影像

Figure 2. 3D model images of Qianguqing scenic spot

下载: 全尺寸图片幻灯片

图 3 漳扎镇小学的三维模型影像

Figure 3. 3D model images of Zhangzha primary school

下载: 全尺寸图片幻灯片

图 4 漳扎镇小学的典型建筑物震害表现

Figure 4. Seismic damage of typical building in Zhangzha primary school

下载: 全尺寸图片幻灯片

图 5 千古情风景区正面的影像分割结果

Figure 5. Segmentation result of the front of buildings of Qianguqing sceic spot

下载: 全尺寸图片幻灯片

图 6 千古情风景区（a）和漳扎镇小学（b）的建筑物提取结果

红色代表完好建筑物面，蓝色代表破坏建筑物面，黑色代表其它地物及背景

Figure 6. Extraction results of seismic damage of buildings in Qianguqing scenic spot （a） and Zhangzha primary school （b）

Red represents the intact building surface，blue represents the damaged building surface and black represents other ground objects and backgrounds

下载: 全尺寸图片幻灯片

表 1 分类样本集的选取

Table 1. Selection of classification sample sets

标签	样本名称	样本图片	样本个数
1	完好建筑物面		107
2	破坏建筑物面		68
3	其它地物及背景		36

下载: 导出CSV

表 2 千古情风景区的建筑物深度学习和目视解译提取结果的混淆矩阵

Table 2. Confusion matrix of deep learning and artificial visual extraction results of the buildings in Qianguqing scenic spot

预测真实	破坏建筑物面数	完好建筑物面数	其它地物及背景面数	合计	正确率
破坏建筑物面数	53 436	20 084	8 007	81 527	65.5%
完好建筑物面数	23 447	131 605	32 258	187 310	70.3%
其它地物及背景数	5 007	19 001	309 974	333 982	92.8%
合计	81 890	170 690	350 239	602 819

下载: 导出CSV

表 3 漳扎镇小学及其周边的深度学习和人工目视解译提取结果混淆矩阵

Table 3. Confusion matrix of deep learning and manual visual extraction results of the buildings in Zhangzha primary school and it’s vicinity

预测真实	破坏建筑物面数	完好建筑物面数	其它地物及背景面数	合计	正确率
破坏建筑物面数	61 445	20 007	5 003	86 445	71.1%
完好建筑物面数	3 668	165 570	44 093	213 331	77.6%
其它地物及背景数	4 008	96 971	688 942	789 921	87.2%
合计	69 121	282 548	738 038	1 089 697

下载: 导出CSV

参考文献(28)

陈梦,王晓青. 2019. 全卷积神经网络在建筑物震害遥感提取中的应用研究[J]. 震灾防御技术,14(4):810–820. doi: 10.11899/zzfy20190412

Chen M,Wang X Q. 2019. The study on extraction of seismic damage of buildings from remote sensing image based on fully convolutional neural network[J]. Technology for Earthquake Disaster Prevention,14(4):810–820 (in Chinese).

范荣双,陈洋,徐启恒,王竞雪. 2019. 基于深度学习的高分辨率遥感影像建筑物提取方法[J]. 测绘学报,48(1):34–41. doi: 10.11947/j.AGCS.2019.20170638

Fan R S,Chen Y,Xu Q H,Wang J X. 2019. A high-resolution remote sensing image building extraction method based on deep learning[J]. Acta Geodaetica et Cartographica Sinica,48(1):34–41 (in Chinese).

冯丽英. 2017. 基于深度学习技术的高分辨率遥感影像建设用地信息提取研究[D]. 杭州: 浙江大学: 22–39.

Feng L Y. 2017. Research on Construction Land Information Extraction From High Resolution Images With Deep Learning Technology[D]. Hangzhou: Zhejiang University: 22–39 （in Chinese）.

高扬. 2018. 基于卷积神经网络的高分辨率遥感影像建筑物提取[D]. 南京: 南京大学: 37–51.

Gao Y. 2018. Building Extraction From High Resolution Remote Sensing Images Based on Convolutional Neural Network[D]. Nanjing: Nanjing University: 37–51 （in Chinese）.

荆帅军,帅向华,甄盟. 2019. 基于无人机倾斜影像的三维建筑物震害精细信息提取[J]. 地震学报,41(3):366–376.

Jing S J,Shuai X H,Zhen M. 2019. Fine information extraction of 3D building seismic damage based on unmanned aerial vehicle oblique images[J]. Acta Seismologica Sinica,41(3):366–376 (in Chinese).

李强. 2018. 多模式遥感数据地震应急关键技术研究[D]. 哈尔滨: 中国地震局工程力学研究所: 33–49.

Li Q. 2018. Study on Key Technology of Earthquake Emergency Using Multi-Mode Remote Sensing Data[D]. Harbin: Institute of Engineering Mechanics, China Earthquake Administration: 33–49 （in Chinese）.

李胜军. 2013. 基于倾斜航空影像的建筑物结构性损毁评估方法研究[D]. 成都: 西南交通大学: 12–20.

Li S J. 2013. The Research of Building Structural Damage Assessment Based on Airborne Oblique Images[D]. Chengdu: Southwest Jiaotong University: 12–20 （in Chinese）.

史路路. 2018. 基于卷积神经网络的遥感影像土地覆盖分类研究[D]. 北京: 中国科学院遥感与数字地球研究所: 14–18.

Shi L L. 2018. Remote Sensing Images Land Cover Classification Based on Convolutional Neural Network[D]. Beijing: Institute of Remote Sensing and Digital Earth, Chinese Academy of Sciences: 14–18 （in Chinese）.

吴剑,陈鹏,刘耀林,王静. 2013. 震害损毁建筑物高分辨率遥感信息提取方法[J]. 地理与地理信息科学,29(3):35–38.

Wu J,Chen P,Liu Y L,Wang J. 2013. Method of earthquake collapsed building information extraction based on high-resolution remote sensing[J]. Geography and Geo-Information Science,29(3):35–38 (in Chinese).

赵妍,张景发,姚磊华. 2016. 基于面向对象的高分辨率遥感建筑物震害信息提取与评估[J]. 地震学报,38(6):942–951. doi: 10.11939/jass.2016.06.014

Zhao Y,Zhang J F,Yao L H. 2016. Seismic damage information extraction and evaluation of buildings with high resolution remote sensing based on object-oriented method[J]. Acta Seismologica Sinica,38(6):942–951 (in Chinese).

周阳,张云生,陈斯飏,邹峥嵘,朱耀晨,赵芮雪. 2019. 基于DCNN特征的建筑物震害损毁区域检测[J]. 国土资源遥感,31(2):44–50.

Zhou Y,Zhang Y S,Chen S Y,Zou Z R,Zhu Y C,Zhao R X. 2019. Disaster damage detection in building areas based on DCNN features[J]. Remote Sensing for Land and Resources,31(2):44–50 (in Chinese).

Gerke M,Kerle N. 2011. Automatic structural seismic damage assessment with airborne oblique pictometry^© imagery[J]. Photogramm Eng Remote Sens,77(9):885–898. doi: 10.14358/PERS.77.9.885

Hu Z W,Li Q Q,Zou Q,Zhang Q,Wu G F. 2016. A bilevel scale-sets model for hierarchical representation of large remote sensing images[J]. IEEE Trans Geosci Remote Sens,54(12):7366–7377. doi: 10.1109/TGRS.2016.2600636

Lecun Y,Bottou L,Bengio Y,Haffner P. 1998. Gradient-based learning applied to document recognition[J]. Proc IEEE,86(11):2278–2324. doi: 10.1109/5.726791

Sun G Y,Huang H,Zhang A Z,Li F,Zhao H M,Fu H. 2019. Fusion of multiscale convolutional neural networks for building extraction in very high-resolution images[J]. Remote Sens,11(3):227. doi: 10.3390/rs11030227

Vetrivel A,Gerke M,Kerle N,Nex F,Vosselman G. 2018. Disaster damage detection through synergistic use of deep learning and 3D point cloud features derived from very high resolution oblique aerial images,and multiple-kernel-learning[J]. ISPRS J Photogramm Remote Sens,140:45–59.

Yuan J Y. 2016. Automatic building extraction in aerial scenes using convolutional networks[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence,40(11):2793–2798.

施引文献

资源附件(0)

访问统计