Key points are not available for this paper at this time.
Accurate identification of complex semi-rigid base cracks (e.g., transverse, longitudinal, etc.) in asphalt pavement using two-dimensional ground penetrating radar (GPR) remains challenging due to unclear edges and particle-filled morphologies. To tackle this problem, an innovative method for the joint identification of multi-profile GPR echo features of base cracks has been proposed. Firstly, three-dimensional GPR (3D-GPR) as a non-destructive testing (NDT) technology, was used to scan pavement internal structure efficiently. Then, the pavement vertical and horizontal sections GPR images containing echo features of base cracks (Crack-V and Crack-H), were obtained to construct deep learning datasets (training, validation, and test sets). The datasets have 1220 pavement GPR images, including 2150 and 1111 echo features of base cracks from vertical, horizontal sections GPR images, respectively. Finally, the networks of YOLOv5 series models were trained, validated, and tested by these datasets sequentially. The results show that when the recognition results of Crack-V and Crack-H based on YOLOv5x model are combined for judgment, the overall detection accuracy of base cracks can reach 93.1%, which is a 5.6% improvement compared to that of a single-dimensional GPR profile. The proposed method can effectively detect both transverse and longitudinal cracks in the base layers.
Xiong et al. (Mon,) studied this question.