地铁隧道病害检测深度学习模型优化及应用  

Optimization and application of deep learning model-based subway tunnel defect detection

作　　者：尤相骏 赵霞[2] 龙四春[3] 王嘉伟 郑颖 邝利军 YOU Xiangjun;ZHAO Xia;LONG Sichun;WANG Jiawei;ZHENG Ying;KUANG Lijun(Zhejiang Huazhan Institute of Design Co.,Ltd.,Ningbo 315000,China;School of Artificial Intelligence,Beijing University of Technology and Business,Beijing 102446,China;School of Earth Science and Space Information Engineering,Hunan University of Science and Technology,Xiangtan 411201,China;China Construction Fifth Engineering Bureau Co.,Ltd.,Shenzhen 518108,China)

机构地区：[1]浙江华展研究设计院股份有限公司,浙江宁波315000 [2]北京工商大学计算机与人工智能学院,北京102446 [3]湖南科技大学地球科学与空间信息工程学院,湖南湘潭411201 [4]中国建筑第五工程局有限公司,广东深圳518108

出　　处：《测绘通报》2024年第8期96-101,共6页Bulletin of Surveying and Mapping

基　　金：国家自然科学基金(42377453;41877283);湖南省科技创新计划(2021RC4037;2033JJ30235);湖南省自然资源厅科研项目(2021-18);2024年度宁波市“科创甬江2035”关键技术突破计划科研项目。

摘　　要：本文针对地铁隧道中渗漏水、裂纹裂缝、结构抹灰开裂及剥落掉块4种常见病害,研究了基于激光雷达扫描点云数据和深度学习的地铁隧道病害检测方法。首先,引入ACmix注意力模块,使网络兼顾全局特征和局部特征,提升对裂纹、裂缝等小目标的检测效果;然后,优化回归损失函数,提高收敛平稳度和回归精度,降低检测误差;最后,实现正射投影图像预处理、批量检测、结果融合及检测结果报表的一体化生成,提高大尺度正射投影图的病害检测率。试验结果表明,在选取IoU阈值为0.5的条件下,改进后的YOLOv8算法在隧道病害测试检测中正确率由90.65%提升至91.18%,基本实现了基于激光雷达扫描的地铁隧道4类常见病害的智能检测,并在实际隧道运维工程中得到成功应用。Aiming at the four common defects of subway tunnel,such as leakage,crack,structural plaster cracking and spalling,a defect detection method of subway tunnel based on laser radar scanning point cloud data and deep learning is studied.Firstly,the ACmix attention module is introduced into the YOLOv8 model to make the network take into account both global and local features,and improve the detection effect of small targets such as cracks and cracks.Then,the regression loss function is optimized,the convergence stability and regression accuracy are improved,and the detection error is reduced.Finally,the complete process of orthographic projection image preprocessing,batch detection and result fusion,and report generation of detection results is realized,and the defect detection of large-scale orthographic projection is efficiently realized.The experimental results show that under the condition that the IoU threshold is 0.5,the mAP of the improved YOLOv8 algorithm on the tunnel defect test set increases from 90.65%to 91.18%,and the AP of cracks increases from 77.89%to 78.70%.The intelligent detection of four common defects of subway tunnel based on LiDAR scanning is solved,and has been successfully applied in actual tunnel operation and maintenance engineering.

关 键 词：深度学习 模型优化 检测方法 隧道病害 

分 类 号：P237[天文地球—摄影测量与遥感]