ABSTRACT
Ground penetrating radar (GPR) is a non-destructive testing method that uses high-frequency electromagnetic waves to detect subsurface defects. GPR offers an extensive detection range and short acquisition time, making it well-suited for assessing the condition of backfill grout behind segment lining of shield TBMs. However, the TBM segment lining with reinforcing bars creates clutter in GPR images due to the shielding effect of the rebars, which obscures the signals of defects in the grout and complicates reliable evaluation. To address this issue, this study developed a rebar clutter elimination model using deep learning-based image segmentation techniques to remove rebar-induced clutter in GPR images and enhance the visibility of defect signals. A database was constructed using GPR numerical simulations based on the finite-difference time-domain (FDTD) method, and various image segmentation techniques, such as Fully Convolutional Network (FCN) and Deeplab V3+, were applied. The clutter removal performance of each method was compared using three image similarity metrics: peak signal-to-noise ratio (PSNR), structural similarity index measure (SSIM), and multi-scale SSIM (MS-SSIM). The results indicated that the Deeplab V3+ method outperformed the FCN method, filtering out clutter with greater clarity. The proposed model is expected to improve the identification rate of defects in the backfill grout behind segment linings and significantly enhance the performance of the segment maintenance system.
