基于改进 U-Net 算法的预制纤维混凝土与普通混凝土接合面图像分割研究

作者：关宏波

作者简介：

关宏波,高级工程师,E-mail:704655705@ qq. com

作者单位：

中铁铁工城市建设有限公司,山东济南 250002

基金项目：

∗江苏省自然科学基金(BK20231429)

摘要：

针对预制装配式混凝土构件接合面质量检测难题,提出一种改进 U-Net 图像分割算法,通过引入Transformer 架构的全局语义建模能力,结合 U-Net 模型局部特征提取优势,构建了面向纤维混凝土和普通混凝土接合面破坏面图像特征的图像分割模型。试验结果表明,U-Net∗模型在真实接合面数据集上平均交并比(mIoU)达78.15%,较传统 U-Net 模型提高 6.45%,改进算法设计提升了 U-Net 网络对接合面图像的识别精确度,证明了改进算法在预制纤维混凝土与普通混凝土接合面识别分割上的应用价值。

关键词：装配式;混凝土;质量;检测;图像分割

English：

Aiming at the problem of joint surface quality detection of prefabricated concrete components,an improved U-Net image segmentation algorithm is proposed. By introducing the global semanticmodeling ability of Transformer architecture and combining the advantages of U-Net model local featureextraction, an image segmentation model for the failure surface image features of the joint surface of fiberconcrete and ordinary concrete is constructed. The experiment results show that the mean intersectionover union ( mIoU) of the U-Net ∗ model on the real joint surface data set is 78.15%, which is 6.45%higher than that of the traditional U-Net model. The improved algorithm design improves the identificationaccuracy of the joint surface image of the U-Net network, and proves the application value of theimproved algorithm in the identification and segmentation of the joint surface of prefabricated fiberconcrete and ordinary concrete.

[1]	吴刚, 冯德成, 徐照, 等. 装配式混凝土结构体系研究进展[J]. 土木工程与管理学报,2021,38(4):41-51,77.WU G,FENG D C,XU Z,et al. Research developments in precastconcrete structural systems[ J]. Journal of civil engineering andmanagement,2021,38(4):41-51,77.
[2]	KHAN M I,ALI AL-OSTA M,AHMAD S,et al. Seismic behaviorof beam-column joints strengthened with ultra-high performancefiber reinforced concrete [ J]. Composite structures, 2018, 200:103-119.
[3]	安晓鹏,李清海,武斌. 纤维增强混凝土连接的装配式混凝土结构节点性能综述 [ J ]. 硅酸盐通报, 2020, 39 ( 11 ):3399-3406.AN X P, LI Q H, WU B. Performance of joints for precastconcrete elements using cast-in-place fiber reinforcedconcrete[J]. Bulletin of the Chinese ceramic society, 2020,39(11):3399-3406.
[4]	刘雷,孙志娟,刘继良,等. 带竖向接缝装配式剪力墙易损性研究[J]. 工程力学,2024,41(S1):150-156.LIU L, SUN Z J, LIU J L, et al. Fragility curves of precastconcrete shear walls with vertical joints [ J ]. Engineeringmechanics,2024,41(S1):150-156.
[5]	淡浩,吴体,肖承波,等. 预制混凝土构件结合面抗剪性能试验及分析[J]. 建筑科学,2020,36(3):100-104.DAN H, WU T, XIAO C B, et al. Test and analysis on shearbehavior of precast component interface [ J]. Building science,2020,36(3):100-104.
[6]	吴香国,张孝臣. 预制超高性能混凝土修复层与既有混凝土界面黏结短期性能研究[ J]. 建筑结构学报,2018,39( 10):156-163.WU X G, ZHANG X C. Investigation of short-term interfacialbond behavior between existing concrete and precast ultra-highperformance concrete layer [ J]. Journal of building structures,2018,39(10):156-163.
[7]	TAYEH B A,ABU BAKAR B H,MEGAT JOHARI M A,et al.Mechanical and permeability properties of the interface betweennormal concrete substrate and ultra high performance fiberconcrete overlay[ J]. Construction and building materials,2012,36:538-548.
[8]	王溧浤,孙新建. 基于 DIC 技术的混凝土剪切力学性能研究[J]. 混凝土,2025(3):6-11.WANG L H,SUN X J. Research on shear mechanical propertiesof concrete based on DIC technology [ J]. Concrete,2025 ( 3):6-11.
[9]	苏怀智,张林海,佟剑杰. 混凝土结构损伤声发射检测进展[J]. 无损检测,2015,37(1):90-95.SU H Z,ZHANG L H,TONG J J. Review on acoustic emissiondetection for concrete damage [ J ]. Nondestructive testingtechnologying,2015,37(1):90-95.
[10]	GU J X, WANG Z H, KUEN J, et al. Recent advances inconvolutional neural networks[ J]. Pattern recognition,2018,77:354-377.
[11]	HARALICK R M, SHAPIRO L G. Image segmentationtechniques [ C] / / Proceedings of the Conference on ComputerVision, Graphics, and Image Processing,1985.
[12]	RONNEBERGER O,FISCHER P,BROX T. U-net:convolutionalnetworks for biomedical image segmentation[M]. Cham:SpringerInternational Publishing,2015.
[13]	SI C Y, YU W H, ZHOU P, et al. Inception transformer[M].Cambridge: MIT Press,2022.
[14]	DOSOVITSKIY A, BEYER L, KOLESNIKOV A, et al. Animage is worth 16×16words: transformers for image recognition atscale[Z]. 2020.
[15]	VASWANI A, SHAZEER N, PARMAR N, et al. Attention isall you need [ C] / / Advances in Neural Information ProcessingSystems,2017.
[16]	SU J H,CHEN Z H, HE C H, et al. GSENet: global semanticenhancement network for lane detection[C] / / Proceedings of theAAAI Conference on Artificial Intelligence,2024.
[17]	LI B N,HU Y H,NIE X C,et al. DropKey for vision transformer[ C] / / 2023 IEEE/ CVF Conference on Computer Vision andPattern Recognition (CVPR),2023.
[18]	HE K M,ZHANG X Y,REN S Q,et al. Deep residual learning forimage recognition [ C ] / / 2016 IEEE Conference on ComputerVision and Pattern Recognition (CVPR),2016.
[19]	RUMELHART D E, HINTON G E, WILLIAMS R J. Learningrepresentations by back-propagating errors [ J ]. Nature, 1986,323(6088):533-536.
[20]	RUMELHART D E, HINTON G E, WILLIAMS R J. Learningrepresentations by back-propagating errors [ J ]. Nature, 1986,323(6088):533-536.
[21]	吴胜男. 多层复合盾构隧道管片混凝土接合面力学及渗透性能研究[D]. 南京:东南大学,2023.WU S N. Study on mechanics and permeability of segmentconcrete joint surface of multi-layer composite shield tunnel[D].Nanjing:Southeast University,2023.
[22]	陈子轩. 预制 UHPC 壳局部增强装配式混凝土框架节点的抗震性能研究[D]. 南京:东南大学,2020.CHEN Z X. Study on seismic performance of precast UHPC shelllocally reinforced precast concrete frame joints [ D]. Nanjing:Southeast University,2020.

试验研究 > 2025年 第卷 第20期