基于多尺度输入图像渗透模型的桥梁裂缝检测

摘要

Aiming at the existing problem of the low precision in crack detection of the surface of railway concrete bridge,a novel bridge crack detection approach based on percolation model with multi-scale input image is proposed.Firstly,weighted piecewise function is employed to enhance contrast ratio,and the optimal threshold segmentation is adopted to largely filter non-crack region.Secondly,different Gaussian kernels are used to get different scales of the input image.Thirdly,multi-scale images of concrete bridge are put into the percolation model to generate high accuracy binary map including only crack information.Finally,the crack information,such as area,length and maximum width,is extracted by the gradients of these cracks on this binary map.Experimental results demonstrate that the proposed approach can improve detection accuracy and stability.%针对现有铁路混凝土桥梁表面裂缝检测方法精确度不高的问题,引入多尺度输入图像渗透模型,提出一种新的桥梁裂缝检测方法.使用加权分段函数进行图像对比度增强,通过最佳阈值分割滤除大部分非裂缝区域,采用不同的高斯核得到不同尺度的输入图像.在渗透模型的基础上,利用多尺度输入图像生成高精度且仅包含裂缝信息的二值裂缝地图,并利用梯度信息提取裂缝的面积、最大宽度及长度等信息.实例验证结果表明,该方法可有效提高检测精确度和稳定性.