Autonomous dynamic line-scan continuous-wave terahertz non-destructive inspection system combined with unsupervised exposure fusion. (December 2022)

• Record Type: Journal Article
• Authors: Hu, Jue; Zhang, Hai; Sfarra, Stefano; Pivarčiová, Elena; Yao, Yuan; Duan, Yuxia; Ibarra-Castanedo, Clemente; Tian, Guiyun; Maldague, Xavier
• Is Part Of: NDT & E international. Volume 132(2022)
• Publication Date: 2022-12
• Subjects: NDT -- Line-scan -- Continuous-wave terahertz -- Infrared thermography -- Unsupervised fusion; Nondestructive testing -- Periodicals; Contrôle non destructif -- Périodiques; Electronic journals; 620.1127
• DOI: 10.1016/j.ndteint.2022.102705
• Languages: English
• ISSNs: 0963-8695

Abstract

Abstract: The investigation of cultural heritage objects is commonly carried out by using non-destructive inspection techniques. In fact, due to the artistic peculiarities of cultural heritage objects, conventional contact-type inspection techniques such as ultrasonic scan cannot always be applied. The objective of this study is to develop a new autonomous dynamic line-scan continuous-wave terahertz (CW THz) non-destructive inspection system combined with long-wave infrared (LWIR) thermography. The newly developed system is aimed at producing clear external and internal maps for wooden objects inherent to the cultural patrimony. Additionally, a new unsupervised fusion algorithm is also proposed for multi-energy density data fusion to correct the inaccurate image acquisition caused by unbalanced line-scan exposure source. The algorithm is designed in an encoder-decoder deep learning structure with dense blocks. Finally, it is worth mentioning that the newly developed system has the capability of speedy detection (up to 49.2 mm/s) and fast in-line processing (of the order of seconds). Highlights: A new autonomous dynamic line-scan non-destructive inspection system combining with continuous-wave terahertz (CW THz) imaging and unsupervised exposure fusion was developed. Line-scanner CW THz technology for inspection was poorly documented in the open literature. The newly developed system is aimed at producing clear external and internal maps for industrial non-destructive inspection. A new unsupervised fusion algorithm UDFF was proposed for multi-focus image fusion to correct the inaccurate image acquisition caused by unbalanced line-scan exposure source. The comparisons exhibit the superiority of UDFF to the state-of-the-art algorithms, especially for background noise suppression and defect details preserving. Experiments and analyses certify the capability of the speedy detection (up to 49.2 mm/s) for the newly developed system.