基于深度学习的倾斜摄影建筑物表面损毁信息提取

Deep Learning-based Extraction of Destruction Information on Building Surface through Oblique Photography

摘要:

建筑物受损信息是地震受灾程度评估的基础,针对传统建筑物表面信息识别人工成本高、效率低等问题,受深度学习提取建筑物影像的启发,提出利用无人机倾斜摄影模型与深度学习相结合的方法提取震后建筑物表面破损信息。以2019年长宁6.0级地震为例,选用双河镇震后倾斜摄影模型切片图为数据源,对比分析面向对象分类方法、VGG-16模型和DeeplabV3+模型对建筑物表面损毁信息的提取结果。分析结果表明,针对建筑物表面破损信息的提取,尤其是细小裂缝的提取,语义分割网络DeeplabV3+模型具有较强的优势(准确率96.93%、召回率96.85%、总体精度96.89%),可实现建筑物表面破损信息的有效提取,具有较强的应用价值。

Abstract:

The destruction information of buildings is essential in earthquake damage assessment. In order to solve the problems of high labor cost and low efficiency in traditional building surface information identification,inspired by extracting building images based on deep learning,we propose a method combining UAV oblique photography model and deep learning to extract the destruction information on building surface after earthquake. In this paper,taking the Changning M_S6.0 earthquake as an example,we selected the slice map of the Shuanghe town oblique photographic model after earthquake as the data source. Then,we conducted a comparative analysis of the object-oriented classification,the network of VGG-16 and DeeplabV3+ for the extraction results of building surface damage information. Our results show that DeeplabV3+ has strong advantages for the extraction of building surface damage information,especially for the small cracks. The accuracy rate,recall rate,and overall precision of the method can reach 96.93%,96.85%,96.89% respectively,which can effectively extract building surface damage information. With more and more sample data accounted,the accuracy rate and recall rate will continue to increase,and the building surface information can be extracted more accurately,which has great practical application value.