| 基于深度学习网络的地震地质灾害识别研究 |
| |
| 引用本文： | 肖本夫,吴今生,毛利,顾铁,梁远玲,代友林.基于深度学习网络的地震地质灾害识别研究[J].震灾防御技术,2021,16(4):617-624. |
| |
| 作者姓名： | 肖本夫 吴今生 毛利 顾铁 梁远玲 代友林 |
| |
| 作者单位： | 四川省地震局, 成都 610041 |
| |
| 基金项目： | 国家重点研发计划（2019YFC1509402）；四川省地震局地震科技专项（LY1817，LY2007）；四川地震科技创新团队专项（201902） |
| |
| 摘 要： | 遥感影像识别方法是破坏性地震震后地质灾害快速、准确获取的重要方法之一，传统的遥感影像识别方法主要以人工目视识别方法和半自动识别方法为主，需投入大量的人力和时间。针对破坏性地震震后地质灾害解译时间长、投入人力多等问题，以2017年8月8日四川九寨沟7.0级地震震后高分辨率无人机遥感影像为研究样本，提出基于深度学习网络的地震地质灾害识别方法。首先结合震后遥感影像解译资料和现场调查资料，提取九寨沟地震地质灾害无人机遥感影像特征，并构建研究区地震地质灾害解译指标和分类数据集；然后采用DeepLabv3+网络结构及softmax损失函数，建立基于深度学习网络的地震地质灾害遥感影像图像语义分割模型方法；最后采用半监督学习方法进行结果验证。研究结果表明，基于深度学习网络的地震地质灾害识别方法可有效识别九寨沟地震地质灾害分布信息，整体分类识别准确率为94.22%，F1分数值为0.77，结果具有较好的一致性和准确性，可提升地震现场灾情获取和重点地震隐患识别等工作效率及服务能力。
|
| 关 键 词： | 深度学习网络 地质灾害 DeepLabv3+ 图像语义分割 九寨沟7.0级地震 |
| 收稿时间： | 2021-10-18 |
Deep Learning Approach for Seismic Geohazard DetectionA Case Study in Jiuzhaigou M7.0 Earthquake,Sichuan, 2017 |
| |
| Xiao Benfu,Wu Jinsheng,Mao Li,Gu Tie,Liang Yuanlin,Dai Youlin.Deep Learning Approach for Seismic Geohazard DetectionA Case Study in Jiuzhaigou M7.0 Earthquake,Sichuan, 2017[J].Technology for Earthquake Disaster Prevention,2021,16(4):617-624. |
| |
| Authors: | Xiao Benfu Wu Jinsheng Mao Li Gu Tie Liang Yuanlin Dai Youlin |
| |
| Institution: | Sichuan Earthquake Agency, Chengdu 610041, China |
| |
| Abstract: | Remote sensing image recognition method is one of the important methods for rapid and accurate acquisition of seismic geohazards after destructive earthquakes. Traditional remote sensing image recognition methods are mainly artificial visual recognition method and semi-automatic recognition method, which need a lot of manpower and time. Aiming at the problems of long time and large investment in the interpretation of geohazards after destructive earthquakes, this paper takes the high-resolution UAV remote sensing images after the Jiuzhaigou M7.0 earthquake in Sichuan on August 8,2017 as the research sample, and proposes a seismic geohazard identification method based on deep learning network. In this study, firstly, the characteristics of UAV remote sensing images of seismic geohazards in Jiuzhaigou were extracted by combining the interpretation data of remote sensing images after the earthquake and the field investigation data, and the interpretation indexes and classification data sets of seismic geohazards in the study area were constructed. Then, the DeepLabv3+ network structure and softmax loss function were used to establish the semantic segmentation model of remote sensing images of seismic geohazards based on deep learning. Finally, the semi-supervised learning method was used to verify the results. The results show that the seismic geohazards identification method based on deep learning can effectively identify the distribution information of Jiuzhaigou seismic geohazards. The overall classification and identification accuracy is 94.22%, and the F1 score is 0.77, indicating that the results have good consistency and accuracy, which can improve the efficiency and service ability of seismic geohazard acquisition and key seismic hazard identification in the earthquake site. |
| |
| Keywords: | Deep learning network Geohazard DeepLabv3+ Semantic image segmentation Jiuzhaigou M< i>7 0 earthquake |
|
| 点击此处可从《震灾防御技术》浏览原始摘要信息 |
| 点击此处可从《震灾防御技术》下载免费的PDF全文 |
|
