摘要
桥梁的定期裂缝检测对于确保桥梁的安全运行至关重要,而目前的人工检测方法不仅耗时费力,还存在诸多不安全因素。而桥梁裂缝还具有多种噪声模式,因此,针对桥梁裂缝的高效检测成为了桥梁健康检测的研究热点和难点。为了实现对较复杂背景下的桥梁裂缝的精确和高效的识别,提出一种结合U-net和Haar-like算法的卷积神经网络的深度学习算法。通过与DenseNet,ResNet,GoogleNet和VGGNet网络的比较,证明了该算法的有效性。同时,该算法还可以对裂缝的面积、长度和平均宽度进行定量计算,检测精度高于97%。研究结果表明:该算法适用于桥梁裂缝图像的高效检测。
The regular crack detection of bridges is very important to ensure the safe operation of the bridge,and the current manual detection methods not only are time consuming and need great effort but also are not very safe.And bridge cracks have multiple noise modes,so the effective detection of bridge cracks is becoming a hot and difficult research topic in bridge health maintenance.In order to achieve accurate and efficient identification of bridge cracks in more complex backgrounds,a deep learning algorithm combined with U-net and Haar-like algorithm was proposed.The comparison with DenseNet,ResNet,GoogleNet and VGGNet proves the effectiveness of the kind of algorithm.The algorithm can also achieve quantitative calculation of the area,length and average width of cracks,and detection accuracy maintains 97%.The results show that the algorithm is suitable for efficient detection of bridge crack images.
作者
杨杰文
章光
陈西江
班亚
YANG Jiewen;ZHANG Guang;CHEN Xijiang;BAN Ya(School of Safety&Emergency Management,Wuhan University of Technology,Wuhan 430079,China;Chongqing Measurement Quality Examination Research Institute,Chongqing 404100,China)
出处
《铁道科学与工程学报》
CAS
CSCD
北大核心
2020年第11期2722-2728,共7页
Journal of Railway Science and Engineering
基金
国家自然基金青年科学基金资助项目(41501502)
重庆市质量技术监督局科研计划项目(CQZJKY2018004)
重庆市技术创新与应用发展专项面上项目(cstc2019jscx-msxmX0051)
长江科学院开放研究基金资助项目(CKWV2019758/KY)。
关键词
桥梁安全
裂缝检测
较复杂背景下
卷积神经网络
深度学习
定量计算
bridge safety
crack detection
under complex background
convolutional neural network
deep learning
quantitative calculation
