基于预训练神经网络的混凝土病害图像识别技术研究  

Research on Concrete Disease Image Recognition Technology Based on Pre-Training Neural Network

作　　者：胡姝婧 HU Shujing(China Railway Construction Group Co.Ltd.,Beijing 100040,China)

机构地区：[1]中铁建设集团有限公司,北京100040

出　　处：《铁道建筑技术》2024年第6期103-106,共4页Railway Construction Technology

基　　金：中铁建设集团有限公司科研计划项目(JC21-15b-2021-1)。

摘　　要：混凝土结构使用过程中可能因多种因素导致不同程度的损伤或缺陷,例如裂缝、渗漏、碳化、钢筋锈蚀、混凝土剥落等。这些病害会对混凝土结构的使用寿命和安全性造成严重影响。传统的人工检测方法存在可靠性低、成本高、受主观性影响大等问题。因此,本研究提出使用预训练神经网络进行混凝土病害图像识别。该方法利用现有的预训练网络,只需导入混凝土结构健康状态下的图像,并激活相应神经网络特征层,即可计算健康状态下大样本的统计阈值。通过比对测试集的统计阈值,能够实现病害图像分类以及异常区域的标记。相较于传统方法,该方法具有精度高、不受人为经验影响、灵活性高、速度快、成本低等优点。During the service life of concrete structures,various factors can lead to different degrees of damage or defects,including cracks,leaks,carbonization,steel corrosion,and concrete spalling,which seriously affect the service life and safety of concrete structures.Due to the low reliability,high cost,and strong subjectivity of manual crack detection methods,this study explores the use of pre-trained neural networks for image recognition of concrete defects.This method utilizes existing pre-trained networks,which only require importing healthy concrete images and activating corresponding neural network feature layers to calculate the statistical thresholds of large samples in healthy conditions.By comparing the statistical thresholds of the test set,it can classify and label the abnormal regions of defect images.Compared with traditional methods,this method has the advantages of high accuracy,low influence of human experience,high flexibility,fast speed,and low cost for image recognition of concrete defects.

关 键 词：混凝土病害 图像识别 预训练神经网络 深度学习 

分 类 号：TU528.1[建筑科学—建筑技术科学]