最小误差准则与脉冲耦合神经网络的裂缝检测

Detection of crack defect based on minimum error and pulse coupled neural networks

表面裂缝检测能够有效判断混凝土桥梁出现的结构性危险。但裂缝特征的多样性、桥梁表面污点引起的图像噪声以及不均匀照明引起的灰度不均等给裂缝检测带来极大的困难。为能够在复杂背景下检测裂缝,分析裂缝图像特征,由脉冲耦合神经网络(pulse coupled neural networks,PCNN)的运行特征和神经元的状态变化分析简化PCNN模型,将简化PCNN模型用于裂缝图像的分割,根据最小误差准则判断PCNN迭代的终止条件,实现了PCNN的裂缝图像自动分割。由圆形度与扁度结合计算区域特征,去除分割后的各种干扰,实现表面裂缝的有效检测。通过敏感度和特异性计算绘制ROC(receiver operating charac-teristics)曲线,比较不同分割方法的曲线特性以评估算法,对实际裂缝图像的处理结果表明了该方法对裂缝图像检测的有效性。

Surface crack detection can effectively judge structure dangers of concrete bridge.But,crack detection becomes very difficult because of variety of crack characters,image noise caused by bridge surface blots and uneven gray scale caused by asymmetric illumination.In order to detect cracks in complicated background,crack image character is analyzed;PCNN model is simplified through analyzing of its running characters and the state change of nerve cells.Crack image is segmented using the simplified PCNN model;the iterative stop condition of the PCNN model is judged with the rule of minimum error,and PCNN crack image segmentation is carried out automatically.The region characters are calculated according to the degrees of flatness and roundness,the interferences after segmentation are removed,and the surface crack effective detection is achieved.ROC curves are drawn using sensitivity and specificity,and the curve characteristics of different detection methods are compared to evaluate the algorithm.Experiment results using the real images of bridge surface show that the proposed crack detection method is effective.