A novel extraction method for weld defect segmentation seeds using ANDM and clustering

• Published in: Insight (Northampton) Vol. 61; no. 12; pp. 706 - 713
• Main Authors: Dang, Changying, Li, Jiansu, Du, Wenhua, Zeng, Zhiqiang, Wang, Rijun
• Format: Journal Article
• Language: English
• Published: The British Institute of Non-Destructive Testing 01.12.2019
• Subjects: Andm; Seed Clustering; Segmentation Seed Extraction; Weld Defect
• ISSN: 1354-2575; 1754-4904
• DOI: 10.1784/insi.2019.61.12.706

To improve the accuracy and reliability in extracting defect segmentation seeds from a weld radiographic testing (RT) image, a novel extraction method (NESS) using clustering and a novel defect detection method (ANDM) that was presented in a previous paper by one of the authors is proposed in this paper. In the proposed NESS, firstly each column of the weld RT image is accurately analysed by ANDM to judge whether or not it really passes through weld defect regions. Most importantly, one or more defect seeds can be acquired if it passes through a defect region. Secondly, all the defect seeds (a defect seed group) of the RT image are extracted by analysing the entire image. Finally, a sorting-based clustering method is proposed to quickly and accurately search for defect segmentation seeds among all the defect seeds, which can solve the problems concerning the difficulty in determining defect segmentation seeds and the heavy calculational burden of defect segmentation. In order to evaluate the performance of the proposed NESS, some clustering and segmentation experiments have been performed. The experimental results reveal that the proposed NESS achieves high accuracy and reliability in extracting defect segmentation seeds from RT images and is helpful in defect segmentation.