摘要
漏磁内检测是目前管道无损检测的重要手段。在进行漏磁无损检测评估中,异常边缘检测是十分重要的环节,异常边缘的精确程度直接影响到后续的反演评估环节。由于数据噪声的存在,使得边缘检测精度大大下降,特别是复杂异常区域。同时,面对庞大的漏磁数据,一般机器学习算法耗时较多。因此,针对漏磁内检测中异常边缘检测问题,本文提出一种基于数据融合的小波变换漏磁异常边缘检测算法。该算法基于小波多尺度变换与分解,将数据层融合、特征层融合以及决策层融合相结合。首先,原始数据经过多色彩空间变换,并将变换结果进行数据融合。然后,融合数据进行小波多尺度变换。其次,针对每一尺度下变换数据进行小波多层分解,并对每一层级进行小波模极大值边缘检测,将边缘检测结果加入到细节分解系数融合中,并重构数据。最终将多尺度下的边缘检测效果进行融合得到最终检测边缘。实验分别在仿真数据集和真实管道数据集上进行,并和其他边缘检测算法,如Sobel、Canny、Roberts、Prewitt、Log进行了比较,实验结果显示,本文提出的异常边缘检测算法效果优于传统边缘检测算法,边缘指标OA高于70%,能够满足实际工程需要。
The inner detection of magnetic flux leakage(MFL)is an important way of nondestructive testing(NDT)in pipeline.For the nondestructive evaluation,anomaly edge detection is an important part because its accuracy directly affects the subsequent inversion process.Due to the data noise,the accuracy of edge detection is greatly reduced,especially for complex anomaly.Meanwhile,general machine learning algorithms take more time to process huge MFL data.To address these problems,an edge detection algorithm based on data fusion and wavelet transform is proposed.This algorithm is based on wavelet multi-scale transformation and decomposition.The data layer fusion,feature layer fusion and decision layer fusion are combined.First,the original data are transformed by multi-color space,and the transformation results are fused together.The fused data are transformed by wavelet multi-scale.Secondly,wavelet multi-level decomposition is executed for each scale data,and the wavelet modulus maximum edge detection is performed for each level.In addition,the edge detection results are combined with detail decomposition coefficient for fusion and reconstruction.Finally,the edge detection results with multiple scales are fused to obtain the final detection edge.Experiments are conducted on the simulated data and the real pipeline dataset,respectively.Results are compared with other edge detection algorithms,including Sobel,Canny,Roberts,Prewitt,and Log.Experimental results demonstrate that the proposed anomaly edge detection method has better performance than traditional edge detection methods.OA evaluation for our method exceeds 70%,which satisfies requirement of practical application.
作者
曹辉
杨理践
刘俊甫
刘斌
Cao Hui;Yang Lijian;Liu Junfu;Liu Bin(Shenyang University of Technology,Shenyang 110870,China;Shenyang University,Shenyang 110044,China)
出处
《仪器仪表学报》
EI
CAS
CSCD
北大核心
2019年第12期71-79,共9页
Chinese Journal of Scientific Instrument
基金
国家自然科学基金(F011403)资助.
关键词
漏磁检测
边缘检测
小波多尺度
数据融合
magnetic flux leakage
edge detection
wavelet multi-scale
data fusion
作者简介
曹辉,2001年于辽宁大学获得学士学位,2009年于东北大学获得硕士学位。现为沈阳大学讲师,主要研究方向为管道检测和图像处理。E-mail:376421002@qq.com;通信作者:杨理践,1984年于哈尔滨工业大学获得硕士学位。现为沈阳工业大学教授、博士生导师,主要研究方向为无损检测、长输油气管道内检测技术及理论。E-mail:yanglijian888@163.com;刘俊甫,2009年于西南石油大学油气储运专业获得学士学位,目前为中石化长输油气管道检测有限公司高级工程师,主要从事管道的内、外检测及评价工作。E-mali:liujunfu86@126.com
