基于深度学习的桥梁表观裂缝检测算法研究

Research on detection algorithm for bridge apparent crack based on deep learning

针对在复杂背景条件下难以直接对桥梁表观裂缝进行检测的问题,文章提出一种基于深度学习的桥梁表观裂缝检测算法。首先利用滑动窗口算法将采集到的桥梁表观裂缝图像切分为小尺寸的桥梁裂缝面元图像和桥梁背景面元图像,并根据对面元图像的分析,提出一种基于Inception网络和残差网络(ResNet)的桥梁裂缝分类模型,用于桥梁裂缝面元和桥梁背景面元的识别;然后结合桥梁裂缝分类模型与滑动窗口算法对桥梁表观裂缝图像进行检测;最后利用数字图像处理技术测量裂缝宽度。结果表明:该文算法对桥梁表观裂缝有超过99%的分类精度,可满足实际工程需要;实现了裂缝的提取并能准确地定位出裂缝在图像中的位置;根据成像原理能测量出裂缝宽度。与传统的深度学习模型相比,该模型拥有更高的执行效率,可用于大规模检测,更易于应用在桥梁健康检测中。

Aiming at the problem that it is difficult to directly detect bridge apparent cracks under complex background conditions,this paper proposes a bridge apparent crack detection algorithm based on deep learning.Firstly,the collected bridge apparent crack image is divided into small-sized bridge crack patches and bridge background patches by sliding window algorithm,and a bridge crack classification model based on Inception network and residual network(ResNet)is proposed according to the analysis of the patches,which is used to identify bridge crack patches and bridge background patches.Then,the bridge crack classification model and sliding window algorithm are combined to detect the apparent crack image of the bridge.Finally,the width of the crack is measured using digital image processing technology.The experimental results show that the algorithm in this paper has more than 99%classification accuracy for the apparent cracks of the bridge,which can meet the actual engineering needs.The extraction of cracks is realized and the position of cracks in the image can be accurately located.The crack width is measured according to the imaging principle.Compared with the traditional deep learning model,the model has higher execution efficiency,can be used for large-scale detection,and is easier to apply in bridge health detection.