Takanori Imai

and 3 more

In recent years, DNN-based inversion of Ground Penetrating Radar (GPR) data has gained significant attention, primarily using simulated data for training due to the lack of ground truth in real-world scenarios. However, models trained on simulation data often perform poorly on real-world data due to domain discrepancies. This study applies an Unsupervised Domain Adaptation (UDA) method to GPR inversion, introducing a domain classifier within the main inversion DNN. By training the main DNN adversarially against the classifier to learn domain-invariant features, the DNN's performance can be improved even without labeled data in the target dataset. Several simulation datasets with varying characteristics were generated to compare the performance of three domain classifiers. The experiments revealed that the proposed simple domain classifier structure, using 3×3 convolutional local alignment, outperformed both pointwise local alignment and global alignment classifiers. Furthermore, the DNN was trained using unlabeled data from real-world roads measured by on-vehicle GPR, and then applied to a bridge deck slab specimen with known design drawings. From both qualitative and quantitative perspectives, the proposed method demonstrates superior performance in permittivity and material estimation. The method enabled clear detection of rebars, and damage such as deterioration and cracks in the concrete slab. This research is expected provide new insights into the efficient subsurface damage detection of road infrastructures.

Takanori Imai

and 1 more

Vehicle-mounted multichannel ground penetrating radar (MC-GPR) facilitates acquisition of volume images of subsurface structures, but inconsistencies arise due to different transmitted wavelets from each antenna. This paper introduces a methodology, Reflectivity-Consistent Sparse Blind Deconvolution (RC-SBD), that estimates transmitted wavelets and stationary clutter for each antenna, thereby calibrating GPR volume images. Previous Sparse Blind Deconvolution was achieved by alternating minimization, producing a single transmitted wavelet and sparse ground reflectivity. RC-SBD utilizes an assumption of subsurface reflectivity smoothness in the horizontal direction, expressed via a total variation regularization term. This allows a wavelet to be derived for each channel. Stationary clutter variables, such as reflection from the vehicle itself and direct waves, are integrated into the model for simultaneous estimation. The objective function incorporates several ℓ2 and ℓ1 regularization terms and is solved using the Split Bregman algorithm augmented by a gradient method. Hyperparameters are determined through Bayesian optimization, aiming to maximize kurtosis of the frequency domain calibrated volume image. The proposed methodology was validated with synthetic data, demonstrating accurate wavelets estimation and significant denoising of volume image. Real-world data application revealed substantial enhancements in the channel depth cross section, visualizing responses of structures such as rebar and steel plates. Furthermore, calibrated image remained stable across diverse datasets, including earthwork and bridge sections, indicating the estimates’ versatility.

Tsukasa Mizutani

and 1 more