摘要
裂缝是混凝土结构危险性鉴定的主要内容,严重影响结构的使用性能并会降低结构的使用寿命,准确分割裂缝并监测所处状态具有十分重要的意义。传统的裂缝检测技术效率低、成本高、主观性强。将计算机视觉的新方法应用于裂缝检测成为一项有挑战性的技术问题,通过对裂缝图像特点与现有语义分割方法的分析,针对混凝土裂缝语义分割存在的裂缝多尺度、主干网络的特征提取等问题,选择使用了适当的模型以及主干网络,并在CRACK500测试集中得到67%的IoU,证实了本文方法语义分割在混凝土结构表面裂缝检测中应用的科学性。
Crack is the main content of risk identification of concrete structure, which seriously affects the service performance of the structure and reduces the service life of the structure. It is of great significance to accurately segment the crack and monitor the state. The existing crack detection technology has the advantages of low efficiency, high cost and strong subjectivity. Applying the new method of computer vision to crack detection has become a challenging technical problem. Through the analysis of the characteristics of crack image and existing semantic segmentation methods, aiming at the problems of crack multi-scale and feature extraction of backbone network in concrete crack semantic segmentation, this paper selects and uses appropriate models and backbone network, and obtains 67% IOU in crack500 test set, It is proved that the semantic segmentation method in this paper is scientific in the application of concrete structure surface crack detection.
作者
王剑
袁辉
芮挺
赵启林
尹初
WANG Jian;YUAN Hui;RUI Ting;ZHAO Qilin;YIN Chu(School of Mechanical and Power Engineering,Nanjing Tech University,Nanjing 211816,China;Field Engineering College,Army Engineering University of PLA,Nanjing 210023,China)
出处
《建筑结构》
CSCD
北大核心
2022年第S02期923-929,共7页
Building Structure
关键词
裂缝检测
语义分割
特征提取
主干网络
crack detection
semantic segmentation
feature extraction
backbone