采用小波基神经网络进行埋地管道缺陷特征提取

由管道漏磁检测的信号描述出管道缺陷的几何特征一直是管道漏磁检测的重点,本文采用小波基神经网络的方法,建立了由管道缺陷的漏磁信号到缺陷截面轮廓图的关系映射.通过ISODATA动态聚类的算法和小波理论中二值扩展的方法选取基函数的中心,经过多层分辨率的训练,网络输出表明该网络可以较准确的反映出缺陷的几何特征,为管道缺陷的特征提取提供了一种可行的方法.

Extracting Characters of Defects in Buried Pipelines Using Wavelet Basis Function Neural Networks

It is the emphasis in pipeline MFL inspections that to de scribe characters of defects in buried pipelines from pipeline MFL inspection si gnals. This paper established a relation mapping from pipeline MFL inspectio n signals to profile of defects using wavelet basis function neural networks met hod in which we select centers of basis functions by ISODATA dynamic clustering algorithm and dyadic expansion scheme. After training this multiresolution wavel et basis function neural network, the output indicated that this net can reflect the characters of defects comparatively exactly, therefore it can be a feasible method to extract the characters of pipeline defects.