基于卷积神经网络的无人机成像桥梁裂缝检测方法研究

Research on Bridge Crack Detection Method of UAV Imaging Based on Convolutional Neural Network

针对桥梁裂缝病害检测困难、裂缝宽度计算精度不高的问题,提出基于卷积神经网络的无人机成像桥梁裂缝检测系统,获取桥梁裂缝图像并提取裂缝,精确计算最大裂缝宽度。该系统改造无人机实现对桥梁底面和侧面图像的采集。首先采用神经网络模型筛选出裂缝图像;然后根据所采集到的图像特点,搭建基于卷积神经网络(Convolutional Neural Network,CNN)的裂缝滑动窗口检测(Slide Crack Detection,SCD)模型,进行图像的小窗口滑动识别以提取裂缝,并与采用基于裂缝图像统计特征的改进中值滤波去噪算法对比裂缝提取效果;最后提出裂缝分类及最大裂缝宽度计算方法,并与裂缝实测结果进行对比。结果表明:该无人机成像桥梁裂缝检测系统对裂缝图像的扰动小,裂缝提取效果更精确,该系统检测并计算的最大裂缝宽度相对实测结果误差在0.05 mm以内,满足桥梁裂缝检测要求。

This paper introduces a bridge crack detection system based on convolutional neural network(CNN),which can collect crack images via unmanned aerial vehicles(UAVs),extract cracks,and accurately work out the maximum width of cracks,thus facilitating crack detection and improving the crack width calculation accuracy.The UAVs of this system are trained to collect images of deck soffit and edges.Firstly,the CNN model is built up to filter deck images,and subsequently,a slide crack detection(SCD)model that is based on CNN is established considering the characteristics of these selected images,to extract images through moving of small sliding windows,and the image extraction efficiency was compared with that of an improved median filtering image denoising algorithm that is developed on the basis of crack image statistical features.The cracks are classified and the maximum crack widths are calculated and the calculated results are compared with field measurements.It is shown that the proposed bridge crack detection system of UAV imaging slightly disturbs crack images and displays more efficient and accurate crack extraction.The errors between the maximum crack widths calculated by the system and the measured results are less than 0.05 mm,meeting the required detection levels.