基于深度学习的隔震构造节点检测方法  

Detection method for isolation details based on deep learning

作　　者：党育[1] 陈一杰 贺一哲 何亚 Dang Yu;Chen Yijie;He Yizhe;He Ya(School of Civil Engineering,Lanzhou University of Technology,Lanzhou 730050,China)

机构地区：[1]兰州理工大学土木工程学院,兰州730050

出　　处：《东南大学学报（自然科学版）》2024年第4期944-951,共8页Journal of Southeast University：Natural Science Edition

基　　金：国家自然科学基金资助项目(51668043,62166025);甘肃省重点研发计划资助项目(21YF5GA073)。

摘　　要：为提高隔震构造节点的检测效率和效果并实现其自动化检测,按构造做法将隔震构造节点分为设置水平隔震缝构造节点、设置竖向隔震缝构造节点和管线设置柔性连接构造节点3类.选取设置水平隔震缝的隔震构造节点,根据其检测要求和缺陷特征,提出了一种基于深度学习的隔震构造节点检测方法.通过实地拍摄国内128个隔震工程的隔震构造节点图像,建立并标定了隔震构造节点数据集,将深度残差网络模型与迁移学习技术相结合,构建了隔震构造节点检测模型.结果表明,该模型在测试集上的识别准确率达到98.4%,F_(1)分值为0.984,可较好地提取水平隔震缝特征,准确判断隔震构造节点是否存在水平隔震缝缺陷,从而实现设置水平隔震缝构造节点的自动检测.To improve the detection efficiency and effectiveness and achieve automated detection of isolation details,three types including the isolation details with horizontal isolation seams,the ones with vertical isolation seams and the ones whose pipeline were with flexible joints were classified according to construction details.The isolation details with horizontal isolation seams were selected.Based on the detection requirements and defect characteristics,a deep learning-based method for detecting isolation details was proposed.By taking on-site photos of isolation details of 128 isolated buildings in China,a data set of isolation details was established and calibrated.The isolation details detection model was constructed by combining the residual network model with transfer learning technology.The results show that the recognition accuracy of this model reaches 98.4%and the F_(1)score is 0.984 on the test sets.The proposed model can extract the characteristics of horizontal isolation seams,and accurately determine the presence of horizontal isolation seam defects in the isolation details,and realize automatic detection of the isolation details with horizontal isolation seams.

关 键 词：深度学习 隔震建筑 水平隔震缝 缺陷检测 

分 类 号：TU352.12[建筑科学—结构工程]