欢迎访问《图学学报》 分享到：
图学学报

图学学报

• 视觉与图像 • 上一篇    下一篇

管道缺陷自动检测与分类

  

  1. 南京航空航天大学机电学院，江苏 南京 210016
  • 出版日期:2017-12-30 发布日期:2018-01-11

Automated Pipe Defect Detection and Classification

  1. College of Mechanical and Electrical Engineering, Nanjing University of Aeronautics and Astronautics, Nanjing Jiangsu 210016, China
  • Online:2017-12-30 Published:2018-01-11

PDF (PC)

342

摘要： 管道作为工业、核设施、石油天然气等领域中常用的物料输送手段，在使用过程中
极易出现各类缺陷，传统的人工检测存在准确率低、效率低、成本高等缺点，采用数字图像处理
技术可以对管道图像进行自动检测与分类，有效克服上述缺点。首先使用图像增强、图像分割、
数学形态学以及边界跟踪对图像进行预处理，在提取出缺陷区域的尺寸、形状和纹理特征后，选
择圆形度、凸度、离心率、熵、相关性和聚集度作为模式识别的特征向量，最后综合使用基于粒
子群优化的K-means 聚类分析和统计模式识别分类器进行分类。使用文中的图像预处理算法可以
成功的将管道缺陷提取出来，达到管道缺陷自动检测的目的。基于粒子群优化的K-means 聚类分
析成功的将管道缺陷图像归为裂纹缺陷、管接头缺陷和孔形腐蚀三类，相比于传统K-means 算法，
聚类准确率分别提高9%、16.7%、12.5%。综合使用基于粒子群优化的K-means 聚类分析和统计
模式识别分类器对管道缺陷进行分类，三类缺陷的分类准确率均在80%以上，其中管接头缺陷和
孔形腐蚀的准确率达到90%以上。综上，综合集成出了一套基于数字图像处理技术的管道缺陷自
动检测与分类算法方案，实验结果表明，该算法方案具有自动化程度高、通用性强、准确率高的
特点。

关键词: 管道缺陷检测, 图像处理, 粒子群优化, 聚类分析, 统计模式识别

Abstract: In the fields of industry, nuclear facilities, oil and gas, pipe is commonly used as the means of
material delivery. And it is easy to appear various defects. The traditional manual detection system has
the disadvantages of low accuracy, low efficiency and high cost. The digital image processing
technology can automatically detect and classify the pipe image, thus effectively overcoming the above
shortcomings. First, image enhancement, image segmentation, mathematical morphology and boundary
tracking are used for image preprocessing. Then, after extracting the size, shape and texture features of
the defective area, we choose the circularity, convexity, eccentricity, entropy, correlation and cluster
tendency as the feature vector. Finally, K-means clustering analysis based on particle swarm
optimization and statistical pattern recognition classifier is used for classification. Using the image
preprocessing algorithm in this paper, we can successfully extract the pipe defects and achieve the
purpose of automated pipe defect detection. K-means clustering analysis based on particle swarm
optimization successfully clusters the pipe defect images into three categories which are crack defects,
pipe joint defects and hole corrosion respectively. Compared with the traditional K-means algorithm, K-means clustering analysis based on particle swarm optimization can increase clustering accuracy by 9%,
16.7% and 12.5% respectively. The clustering analysis based on particle swarm optimization and the
statistical pattern recognition classifier is used to classify the pipe defects. The classification accuracy of
the three types of defects is more than 80%. The accuracy of pipe joint defects and hole corrosion is more
than 90%. In summary, an integrated algorithm scheme for automated pipe defect detection and
classification based on digital image processing technology is proposed. The experiments show that the
algorithm scheme has the characteristics of high degree of automation, high versatility and accuracy.

Key words: pipe defect detection, image processing, particle swarm optimization, cluster analysis, statistical pattern recognition