Recognition of Pavement Structural Damage Based on Nondestructive Testing and Computer Vision

Despite the growing body of research on pavement quality assessment in civil and infrastructure engineering, the issue of achieving rapid analysis of nondestructive structural damage remains unresolved. The application of nondestructive testing, such as ground penetrating radar (GPR) in highway engineering, provides opportunities to analyze distress hidden under pavement surfaces. However, the traditional analysis process of GPR image databases mainly relies on technicians’ experience. Furthermore, existing methods for the recognition and identification of internal damage, such as looseness and cracks, require a high technical threshold and a tremendous amount of time, leading to low efficiency and unstable accuracy of the results. To address this issue, this paper developed an automatic identification method by combining nondestructive testing and computer vision for pavement structural damage. First, the GPR images of the internal damage were calibrated by comparing them to the cores drilled onsite. Then, an object detection (OB) algorithm, called YOLOv5, was adapted to construct machine learning models. A total of 1,000 labeled images were fed into the model, which was trained and optimized to learn and identify the images’ features. The result indicated that the developed method achieved high-speed and accurate identification of the location and type of pavement structural damage. The prediction accuracy of the OB model reached 88.2%, and the GPR image dataset of 10 km pavement with 1,000 samples can be processed in 1 min. The contribution of this research facilitates the development of pavement maintenance decision systems to provide a real-time assessment method for onsite pavement structures.