Frequency‐domain wavefield reconstruction inversion of ground‐penetrating radar based on sensitivity analysis

Abstract

AbstractGround‐penetrating radar full‐waveform inversion is a high‐resolution method for inverting permittivity and conductivity; however, the issue of multi‐parameter crosstalk during the inversion process, in which perturbations of permittivity and conductivity can produce nearly identical observed data, poses a challenge. Additionally, full‐waveform inversion is highly nonlinear and computationally expensive. In this study, we conducted a sensitivity analysis based on permittivity and conductivity perturbations, which demonstrates that their sensitivities vary with frequency. Specifically, conductivity perturbation has a larger impact on low‐frequency data, whereas permittivity perturbation increasingly affects high‐frequency data. Based on the sensitivity analysis, we propose a modified stepped inversion strategy to mitigate multi‐parameter crosstalk. We also employed wavefield reconstruction inversion, which relaxes the wave‐equation constraint as a penalty term and, thus, can help us avoid local minima of the objective function. In contrast, full‐waveform inversion is more prone to get stuck owing to mismatches between modelled and measured data during the inversion process. Finally, we tested the proposed approach on crosshole synthetic data, which achieved significant computational savings and higher inversion efficiency for fewer forward simulations. Our results demonstrate that the proposed approach is a promising method for inverting subsurface structures and has the potential for practical applications in the future.