摘要
针对现有混凝土表面裂缝检测方法对不同环境下采集的裂缝图像集检测效果鲁棒性不强的问题,引入基于结构森林的学习框架来提取裂缝边缘,并融合改进的快速渗流算法检测裂缝,以保证检测精确率和效率。使用分段函数对彩色图像进行线性变换以增强裂缝,根据包含裂缝块的局部结构特征及彩色图像积分通道特征,利用结构森林边缘检测器快速提取裂缝边缘,同时结合改进的渗流模型快速渗流边缘并去噪。最后,利用形态学方法,连接较小断裂并填充孔洞。在收集的各类裂缝图像集上的实验结果表明,该算法处理速度快、鲁棒性好,且裂缝提取的精确度优于现有算法。
To improve the robustness of crack detection methods for different concrete surface crack images,this paper utilized structured forest based learning framework to extract crack edge,and merged improved fast percolation algorithm to detect crack,ensuring the precision and efficiency of detection.This approach enhances the crack images by using a linear transform piecewise function to conduct linear transformation for color images.Then,according to the local structured information of crack block and the integral channel features obtained from the crack edge images,the structured forest edge detector is used to extract the crack edge fast,and the improved percolation model is fused to percolate edge fast and denoise.Finally,the morphological method is used to connect small fractures and fill the holes.Experimental results on various crack image datasets show that the proposed approach is fast and robust,and it’s superior to state-of-the-art algorithms in terms of the accuracy of crack detection.
作者
瞿中
鞠芳蓉
陈思琪
QU Zhong;JU Fang-rong;CHEN Si-qi(School of Software Engineering,Chongqing University of Posts&Telecommunications,Chongqing 400065,China)
出处
《计算机科学》
CSCD
北大核心
2018年第11期288-291,311,共5页
Computer Science
基金
重庆市基础科学与前沿技术研究项目(cstc2015jcyjBX0090
cstc2014jcyjA40033
cstc2015jcyjA40034
cstc2014jcyjA10051)资助
关键词
结构森林
边缘检测
渗流模型
去噪
Structured forest
Edge detection
Percolation model
De-noising
