一种改进型的PCNN路面裂缝检测方法

路面病害的视觉自动检测一直是公路养护研究的热点问题,针对传统脉冲耦合神经网络(PCNN)在路面图像分割中的参数选择以及迭代中最佳分割结果的选择问题,提出通过简化脉冲耦合神经网络的模型,对关键参数自适应以及神经元点火条件的改进,以图中最大的裂缝作外接矩形,在迭代去噪过程中,外接矩形面积大大减小或矩形内部出现更多裂缝,表明出现过分割,利用该特征可以确定迭代中止,取上一次迭代结果,并用连通域分割去除细小噪声。通过受试者工作特征(ROC)曲线分析,该算法在裂缝检测上比基于最大熵的脉冲耦合神经网络分割算法提高8.4%,比基于灰度迭代阈值算法的脉冲耦合神经网络算法提高14.7%,提高了裂缝检测的准确率和抗噪性。

Improved PCNN Pavement for the Pavement Crack Detection Method

The visual automatic detection of pavement diseases has always been a hot issue in highway maintenance research.The paper proposed a method to simplify the pulse-coupled neural network by selecting the parameters of traditional pulse coupled neural network(PCNN)in road image segmentation and the best segmentation results in iteration.The adaptation of key parameters and neuron ignition conditions were improved,and the circumscribed rectangle was the largest crack in the figure.In the iterative denoising process,the area of the circumscribed rectangle was greatly reduced or more cracks appeared inside the rectangle,indicating that there has been over-segmentation.Using this feature to determine the iteration abort,the last iteration result was take.,and the connected domain segmentation was used to remove the small noise.Through the receiver operating characteristic(ROC)curve analysis,the algorithm improves the crack detection by 8.4% compared with the maximum entropy-based pulse coupled neural network segmentation algorithm,and by 14.7% compared with the pulse coupled neural network algorithm based on the grayscale iterative threshold algorithm,which can improve the accuracy and noise immunity of crack detection.