摘要
针对目前裂缝识别存在样本较少,识别精度受样本采集时的环境因素影响较大等问题,提出一种结合VGG网络和Seg Net网络的裂缝分割算法模型(DeepCrack),解决了模型鲁棒性较差的缺陷,实现了裂缝的像素级(pixel-level)识别定位。基于该模型与另外6种深度学习模型在公开道路数据集CRKWH100和CrackL315上的测试结果表明:该模型不仅可以实现对裂缝的识别定位,还能准确地提取裂缝的尺度信息,研究结果可应用于实际工程检测。
To solve the sample-lacking problems of crack identification and the possible low recognition accuracy due to environmental factors during data collection, a DeepCrack algorithm model combining VGG network and Seg Net network is proposed to improve the robustness of the model and realize pixel-level recognition and localization of cracks. Through comparison with different deep learning models on the public road datasets CRKWH100 and CrackL315, the results show that this model can not only realize the identification and localization of cracks, but also accurately extract the scale information of cracks, and the network can be applied to actual engineering inspection.
作者
隆星
LONG Xing(China Railway Construction Investment Group Corp,Ltd.,Zhuhai 519031,China;School of Human Settlements and Civil Engineering,Xi'an Jiaotong University,Xi'an 710049,China)
出处
《工程建设与设计》
2023年第2期105-109,共5页
Construction & Design for Engineering
基金
中国铁建投资集团科技研发项目(ZTJ2021WBXKYKT)。
关键词
高速公路
深度学习
病害识别
卷积神经网络
highway
deep learning
distress identification
convolutional neural network
