AR-SMFWI: A Hybrid Artifact Removal and Spatial Multiscale FWI Approach for Common-Offset GPR Data

Full waveform inversion (FWI) has shown great potential for high-resolution imaging in ground-penetrating radar (GPR) applications. However, existing research has primarily focused on multi-offset or crosshole GPR configurations, while common-offset GPR (Co-GPR) despite its advantages in system simplicity and cost-effectiveness, is less studied due to its inherent ill-posedness and limited angular illumination. To tackle the challenges of deep gradient attenuation, artifact contamination, and high computational cost in Co-GPR FWI, this study introduces a robust and efficient inversion framework. First, an illumination compensation strategy based on envelope extrema is proposed to adaptively enhance sensitivity to deep reflectors, effectively counteracting energy decay without relying on Hessian approximations. Second, a singular value decomposition (SVD)-based filtering method is integrated into the gradient update to suppress artifacts from hyperbolic tails and spurious cross-correlation effects. Furthermore, a frequency-guided spatial multiscale strategy is incorporated within a subspace FWI framework, allowing dynamic adjustment of spatial and temporal resolution according to the dominant frequency, which significantly reduces computational demands. Numerical experiments on synthetic and field 2D/3D data validate the proposed methodology. The results demonstrate that the method produces structurally consistent images with fewer artifacts and enhanced depth resolution. Quantitatively, it achieves a structural similarity index measure (SSIM) of up to 95.67% and a relative SSIM (R-SSIM) exceeding 96%. The proposed approach provides a viable solution for large-scale Co-GPR inversion under limited computational resources and offers new insights for extending FWI to practical GPR imaging scenarios.