基于灰度-梯度共生矩阵的焊缝缺陷聚类分析  

Clustering Analysis of Weld Defect Based on Gray-gradient Co-occurrence Matrix

作　　者：戴光[1] 崔巍[1] 张颖[1] 王学增[2] 

机构地区：[1]东北石油大学机械科学与工程学院 [2]中国石油大庆石化分公司,黑龙江大庆163000

出　　处：《中国安全科学学报》2013年第3期79-85,共7页China Safety Science Journal

基　　金：黑龙江省教育厅科学技术研究重点项目(2511008);黑龙江省研究生创新科研项目(YJSCX2012-048HLJ)

摘　　要：为识别焊缝不同类型缺陷,以焊缝缺陷的漏磁检测(MFL)图像为研究对象,提出基于灰度-梯度共生矩阵(GGCM)和聚类分析的焊缝缺陷识别方法。将采集的三维MFL信号转换为二维灰度图像,利用GGCM直接提取焊缝3种状态下(焊缝无缺陷、焊道上分布圆柱体缺陷、焊道上分布矩形槽缺陷)MFL图像的特征信息。结合2种聚类算法,用层次聚类法选取评述焊缝图像的特征量,用k-均值聚类方法分析这些特征量,并以可视化图形显示聚类结果。结果表明:根据GGCM提取的特征量的聚类分析结果,焊缝典型缺陷的识别率大于96%。In order to identify the different types taken as the research object, a weld defect image was worked out. First, three-dimensional MFL of weld defects, the MFL imaging of the weld defect was recognition method based on GGCM and cluster analysis signal was converted to a two-dimensional gray-scale image. And then MFL image characteristics of the three weld kinds （weld non-defect, cylinder defects in the weld, rectangular slot defects in the weld） were extracted using GGCM. Also, both clustering algo- rithms were combined. Characteristic quantities of weld image were selected using the hierarchical cluste- ring method. Then, the characteristic quantities were analyzed using the k-means clustering. And cluste- ring results were showed as visualization graphics. The outcomes indicate that according to the resuhs of cluster analysis of the feature quantities extracted by GGCM, the recognition rate of the typical weld defects is higher than 96%.

关 键 词：焊缝缺陷 漏磁检测(MFL)图像 灰度-梯度共生矩阵(GGCM) 层次聚类法 k均值聚类法 

分 类 号：X941[环境科学与工程—安全科学]