Paper Info
Title
A Defect Recognition Method for Low-Quality Weld Image Based on Consistent Multiscale Feature Mapping
Abstract
Defect recognition is an essential technology for the safety of pipelines. Recently, deep-learning (DL)-based methods have shown great progress in recognizing pipeline weld defects. However, when it comes to low-quality images in some complicated applications, the DL-based methods can hardly achieve satisfactory results. To solve this problem, this article proposes a consistent multiscale feature mapping method for defect recognition of low-quality weld images. First, a multiscale feature mapping method is proposed, so that different types of defect images can be mapped into distinctive features in both local and global embedding spaces. Second, a novel consistency strategy is proposed to maintain the consistency of the local and global embedding spaces, so that similar defects can be easily distinguished. Third, a special feature fusion model is proposed to fuse the mapped local and global features, so that the performance of defect recognition can be improved. Finally, five groups of experiments are conducted using pipeline weld X-ray images and metal weld X-ray images. The experiment results show that the proposed method is effective in recognizing the weld defects of low-quality X-ray images.
Year
DOI
Venue
2022
10.1109/TIM.2022.3171609
IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT
Keywords
DocType
Volume
Feature extraction, Image recognition, Welding, X-ray imaging, Pipelines, Measurement, Deep learning, Consistency strategy, feature fusion, low-quality X-ray image, multiscale feature mapping, weld defect recognition
Journal
71
ISSN
Citations 
PageRank 
0018-9456
0
0.34
References 
Authors
0
5
Name
Order
Citations
PageRank
Jinhai Liu1135.08
Xiaoyuan Liu200.34
Fuming Qu300.34
H Zhang47027358.18
H Zhang57027358.18