Advances of deep learning applications in ground-penetrating radar: A survey

• Published in: Construction & building materials Vol. 258; p. 120371
• Main Authors: Tong, Zheng, Gao, Jie, Yuan, Dongdong
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 20.10.2020; Elsevier
• Subjects: Data processing; Deep learning; Engineering Sciences; Ground-penetrating radar (GPR); Intelligent inspection for civil engineering; Nondestructive testing (NDT); Deep learning; 99-00; Ground-penetrating radar (GPR); Nondestructive testing (NDT); Intelligent inspection for civil engineering; 00-01; Data processing
• ISSN: 0950-0618; 1879-0526
• DOI: 10.1016/j.conbuildmat.2020.120371

•Discussing advances of deep learning applications in GPR.•Discussing the existing issues of deep learning applications in GPR.•Comparing the architectures of deep leaning models exploiting GPR data.•Introducing the foundation of deep learning and GPR. Deep learning has achieved state-of-the-art performance on signal and image processing. Due to the remarkable success, it has been applied in more challenging tasks, such as ground-penetrating radar (GPR) testing in civil engineering. This paper reviews methods involving deep leaning and GPR for civil engineering inspection and provides a classification based on the data types that they exploit. Based on the results of a comparison study, we conclude that methods using A-scan data slightly surpass the models using B- and C-scan data, though C-scan data is maybe the most promising in the further thanks to its complete space information. Two current limitations of deep learning exploiting GPR are its dependence on big data and overconfident decision-making. Therefore, benchmark GPR data sets and cautious deep learning are required.