基于U-Net模型的建筑结构裂缝分割数字化处理方法研究  

Research on Digital Processing Method of Building Structure Crack Segmentation Based on U-Net Model

作　　者：衣伟平 YI Weiping(East China Branch of China Railway Construction Group Co.,Ltd.,Suzhou 215300,Jiangsu,China)

机构地区：[1]中铁建设集团有限公司华东分公司,江苏苏州215300

出　　处：《工程技术研究》2024年第6期19-23,共5页Engineering and Technological Research

摘　　要：混凝土结构裂缝是一种常见的病害问题,由于引起结构开裂的因素较多,且建筑结构物开裂形态各异。因此,针对上述问题,文章提出一种基于U-Net模型的建筑结构裂缝识别方法,实现裂缝识别的数字化与智能化。以U-Net网络为基础,建立了离散性相对较强的训练集与验证集。在下采样和上采样过程中,进行最大池化与反卷积操作,进行架构调整,并通过Dice系数对模型进行评价。结果表明,利用该方法能够高效且较为准确地进行建筑结构裂缝数字化识别,提高结构病害检测效率。Concrete structure crack is a common disease problem.There are many factors causing structural cracking,and the cracking forms of building structures are different.In view of the above problems,this paper proposes a crack identification method of building structures based on U-Net model to realize the digitization and intelligence of crack identification.Based on the U-Net network,a training set and a verification set with relatively strong discreteness were established.In the process of down-sampling and up-sampling,the maximum pooling and deconvolution operations are performed,the architecture is adjusted,and the model is evaluated by Dice coefficient.The results show that this method can be used to identify the cracks of building structures efficiently and accurately,and improve the detection efficiency of structural diseases.

关 键 词：建筑结构裂缝 U-Net模型 数字化处理 

分 类 号：TU755.7[建筑科学—建筑技术科学]