基于PCNN和遗传算法相结合的新型混凝土桥梁裂缝检测方法

摘要

针对混凝土桥梁裂缝对比度低、裂缝图像噪声干扰强等难题,提出了基于脉冲耦合神经网络(PCNN)和遗传算法相结合的混凝土桥梁裂缝检测新算法(GA-PCNN).该算法首先利用遗传算法优化裂缝PCNN模型参数;然后通过改进的最小对数误差适应度函数区分裂缝与背景,当适应度值大小几乎无变化时,停止分割图像;最后通过连通域去噪算法滤除残余噪声,实现裂缝的自动检测.比较GA-PCNN、PCNN和基于熵及动态阈值算法对裂缝图像的分割效果,并绘制PR和ROC曲线评价分割质量,经计算GA-PCNN算法的PR和ROC曲线下面积为90.6％和91.6％,分别高于PCNN算法10.1％和6.8％、基于熵和动态闽值6.5％和6.7％.实验结果表明,GA-PCNN新算法分割效果好且去噪能力强,该算法能准确地提取混凝土桥梁裂缝特征.%Aiming at low contrast and strong noise interference in the crack of concrete bridges,this paper presented a novel crack detection method for concrete bridges based on pulse coupled neural network (PCNN) and genetic algorithm (GA-PC-NN).Firstly,the algorithm used genetic algorithm to optimize parameters of the fracture PCNN model.Then it discriminated fractures and background by the improved minimum logarithmic error adaptation degree function.The algorithm stopped image segmentation when the fitness value was almost unchanged.Finally,it adopted the algorithm of connected domain filter to eliminate the residual noise.Thus,it achieved automatic crack detection.This paper compared segmentation results of crack images by GA-PCNN,PCNN and algorithm based on entropy and dynamic threshold,and drew the PR curve and ROC curve to evaluate the quality of segmentation.Through calculation,area under the PR curve and ROC curve of GA-PCNN algorithm was 90.6％ and 91.6％,respectively higher than PCNN by 10.1％ and 6.8％,and algorithm based on entropy and dynamic threshold by 6.5％ and 6.7％.Results of the experiments indicate that the segmentation effect and the denoising capability of GA-PCNN is apparent,verify that the algorithm can accurately extract the crack characteristics of concrete bridges.