文章编号：1673-0291（2023）03-0010-09DOI：10.11860/j.issn.1673-0291.20220076第47卷第3期2023年6月Vol.47No.3Jun.2023北京交通大学学报JOURNALOFBEIJINGJIAOTONGUNIVERSITY融合改进ResNet-14和RS-Unet模型的混凝土桥梁裂缝识别梁栋，李英俊，张少杰（河北工业大学土木与交通学院，天津300401）摘要：针对噪声影响下的细小混凝土裂缝检测，提出了将改进的深度残差网络（ResNet-14）和基于U形框架的Swin-Unet网络（RevisedSwin-Unet，RS-Unet）相融合的混凝土桥梁裂缝检测识别方法.首先，利用改进的ResNet-14网络对裂缝子块进行识别，去除划痕、剥落等噪声的干扰，并保留裂缝区域；然后，采用RS-Unet网络模型对图像进行像素级分割，完成裂缝特征提取；最后，采用边缘线最短距离法进行宽度计算，并在实验室条件下设计了一套裂缝检测系统用以验证该方法.试验结果表明：在固定拍摄角度和距离的前提下，融合改进的ResNet-14和RS-Unet网络模型对噪声影响下细小混凝土裂缝的识别效果体现出了良好的抗干扰性和准确性，为其应用于实际工程中提供了重要参考作用.关键词：桥梁工程；深度学习；ResNet；裂缝识别；特征提取；宽度测量中图分类号：U446文献标志码：AIdentificationofcracksinconcretebridgesthroughfusingimprovedResNet-14andRS-UnetmodelsLIANGDong,LIYingjun,ZHANGShaojie（SchoolofCivilandTransportationEngineering,HebeiUniversityofTechnology,Tianjin300401,China）Abstract:Toaddressthedetectionoffineconcretecracksundertheinfluenceofnoise,thispaperpro⁃posesafusionmethodthatcombinestheimproveddeepresidualnetwork(ResNet-14)andtheSwin-UnetnetworkbasedonaU-shapedstructure(RevisedSwin-Unet,RS-Unet)forcrackdetectionandrecognitioninconcretebridges.Firstly,theimprovedResNetnetworkisusedtoidentifycracksub-block,eliminatingthenoiseinterferencesuchasscratchandspalling,whilepreservingthefracturearea.Then,theRS-Unetnetworkmodelisutilizedforpixel-levelsegmentationoftheimagestofacilitatethecrackfeatureextraction.Finally,thewidthofthecracksiscalculatedusingtheshortestdistancemethodalongtheedgelines.Tovalidatetheproposedmethod,asetofcrackdetectionsystemisdesignedandtestedunderlaboratoryconditions.Theexperimentalresultsshowthatunderthepremiseoffixedshoot⁃ingangleanddistanc...
