Complex shape subsurface target 3D reconstruction method based on multi-profile ground penetrating radar data

Abstract

Ground Penetrating Radar (GPR) is widely used in the non-destructive testing of engineering structures. However, the 3D morphology of complex shape subsurface targets is often diverse and highly irregular, making it difficult for conventional methods to clearly depict contour boundaries. To address this problem, this paper proposes 3D GPR-TransUNet, a novel 3D fine imaging network model based on multi-profile imaging results. By introducing a 3D multi-scale feature extraction module and a Transformer module, the model fully excavates the target detail features at different scales and establishes global dependencies. This effectively solves the problem of unclear boundary characterization caused by the incomplete extraction of detail features in conventional methods, realizing the fine characterization and accurate size estimation of complex 3D target morphologies. The effectiveness of the proposed method was validated through simulation experiments using a dataset of 3,500 models and practical concrete model tests. Results demonstrate that our method achieves a mean Intersection over Union (mIoU) of 0.8507, precisely restoring the nonlinear boundaries of subsurface targets and significantly outperforming traditional methods. This study provides reliable technical support for the fine detection of complex shape subsurface targets in engineering structures.