Pavement Layer Interface Detection From GPR Data Using Deep Learning

Ground penetrating radar (GPR) is a non-destructive test widely used for capturing high-resolution road profiles. Accurate measurements of pavement layer thickness are crucial for evaluating the integrity of both new and existing pavement constructions. However, the automatic interpretation of pavement layer interfaces needs to be improved. This study introduces an automated method to identify pavement layer interfaces using an image segmentation approach based on U-net models. Using real-world data, we evaluated three variants, U-net, attention U-net, and R2U-net. Notably, the R2U-net model yielded a dice similarity coefficient of 95.962% and a mean intersection over union (IoU) of 95.567% in interface detection.