Machine learning-based detection of transportation infrastructure internal defects using ground-penetrating radar: a state-of-the-art review

Abstract

Abstract Early detection of internal defects is crucial to ensure the long-term performance and stability of transportation infrastructure. Researchers and practitioners have developed various nondestructive testing (NDT) methods for this purpose. Among them, the ground-penetrating radar (GPR) technique has been widely implemented due to its advantages such as large coverage, traffic-speed survey, and rich subsurface information. In addition, machine learning (ML) algorithms have been frequently applied to achieve automatic GPR data interpretations, which are essential for field applications. However, the fundamental concepts, architectures, and appropriate application scenarios of these algorithms are often questionable to practitioners and researchers. This paper presents a state-of-the-art review of ML applications in the internal defect detection of transportation infrastructure using GPR. In particular, pavements and bridges are covered. The basic knowledge of GPR working principles and ML algorithms is documented. The critical features of the ML algorithms for each detection task are presented. The drawbacks that may hinder the application of ML algorithms using GPR are indicated, including the insufficiency of labeled GPR data, unavailability of GPR dataset, customized ML architecture, and field validation. Finally, possible transfer learning, integrated robotic platform, and data fusion with other NDT methods are discussed. This review paper is expected to serve as a reference for practitioners to choose appropriate ML algorithms to detect internal defects in transportation infrastructure using GPR.