摘要
为了更加精确地识别混凝土表观病害,首先收集了包含混凝土一般性病害、风化、露筋和裂缝四种表观病害的图片,利用图像处理技术对图像集进行了扩充;然后建立了深度残差网络模型,得到了混凝土四种表观病害的分类器;最后通过迁移学习对残差网络模型进行优化,得到最优分类结果.结果表明:该基于深度学习的混凝土表观病害分类器可以针对混凝土单个病害图像进行智能分类,经过迁移学习的优化,准确率达到了91.3%,对混凝土破损露筋病害的识别准确度达到了97.6%,可以满足实际工程中混凝土表观病害智能检测的需要.
To more accurately identify the apparent diseases of concrete,pictures of four kinds of apparent diseases of concrete was collected firstly,including general diseases,weathering,rebar exposed and cracks,and image processing technology was used to expand the image set.Then,the deep residual network model was established to get the classifiers of the four apparent diseases of concrete.Finally,the transfer learning was used to optimize the deep residual network model and get the best classification result.Results shows that the established concrete apparent diseases classifier based on deep learning can intelligently classify the images of the single disease of concrete. Through the optimization of the transfer learning, the accuracy rate reaches 91.3%, and the recognition accuracy of concrete damaged rebar exposed diseases reaches 97.6%,which can meet the needs of intelligent detection of concrete apparent diseases in actual projects.
作者
黄彩萍
甘书宽
谭金甲
黄志祥
HUANG Caiping;GAN Shukuan;TAN Jinjia;HUANG Zhixiang(College of Civil Architecture and Environment,Hubei University of Technology,Wuhan 430068,China)
出处
《华中科技大学学报（自然科学版）》
EI
CAS
CSCD
北大核心
2021年第4期96-101,113,共7页
Journal of Huazhong University of Science and Technology(Natural Science Edition)
基金
国家自然科学基金资助项目(51708188)
桥梁结构健康与安全国家重点实验室资助项目(BHSKL19-01-KF)。
关键词
混凝土病害
分类器
深度学习
迁移学习
深度残差网络
concrete diseases
classifier
deep learning
transfer learning
deep residual network
