基于深度学习的钢桥螺栓关键点识别方法
Key Point Identification Method of Steel Bridge Bolt Based on Deep Learning
徐建平 1刘桂芬 1王杨 2程潜2
作者信息
- 1. 杭州市交通运输发展保障中心,浙江杭州 310012
- 2. 中交公路长大桥建设国家工程研究中心有限公司,北京 100088
- 折叠
摘要
为解决钢结构桥梁高强度螺栓数量多、脱落风险高和人工检查效率低的难题,基于深度学习技术研发了一种通过定位螺母(螺栓头)6个角点和1个中心点来识别高强度螺栓关键点的识别方法.首先,通过实际工程拍摄与数据增强方法,构建了公路钢桥大六角头高强度螺栓数据集.然后,设计并搭建了以ResNet50为主干网络的模型,将标注后的训练集转换为热力图并对模型进行训练,进而提出了钢桥节点螺栓编号规则与算法.最后,以正确关键点百分比与准确率为评估指标对训练得到的模型性能进行了评估,利用新采集的螺栓图像对模型进行关键点定位试验和不同光线下鲁棒性试验,并结合实际工程对关键点的识别精度进行了验证.研究结果表明:室内试验和实际工程中模型螺栓的识别率均为100％,且现场识别效果优于试验结果.该研究成果可为钢桥高强度螺栓病害智能检测提供参考.
Abstract
To solve the problems of a large number of high-strength bolts,high risk of detachment,and low efficiency of manual inspection in steel structure bridges,a recognition method based on deep learning technology has been developed to identify the key points of high-strength bolts by locating 6 corner points and 1 center point of the nut(bolt head).Firstly,a dataset of high-strength bolts with large hexagonal heads for highway steel bridges was constructed through actual engineering photography and data augmentation methods.Then,a network model with the backbone of ResNet50 was designed and built.The annotated training set was converted into a heatmap and the model was trained.Subsequently,a steel bridge node bolt numbering rule and algorithm were proposed.Finally,the performance of the trained model was evaluated by the evaluation indicators of percentage of correct key points and accuracy.Key point localization experiments and robustness tests under different lightings were conducted on the model by newly collected bolt images.And the recognition accuracy of key points was verified through practical engineering.The research results indicate that the recognition rate of model bolts in both indoor experiments and actual engineering are 100％.The on-site recognition effect is better than the experimental results.This research result can provide reference for intelligent detection of high-strength bolt diseases in steel bridges.
关键词
公路桥梁/钢结构/高强度螺栓/深度学习/关键点定位Key words
highway bridges/steel structure/high-strength bolts/deep learning/key point localization引用本文复制引用
出版年
2024
国家科技期刊平台
NSTL
万方数据