MAPD-Net: A GPR-Based Method for Estimating Rebar Parameters and Concrete Moisture Content

Ground-penetrating radar (GPR) is an efficient non-destructive technique for inspecting reinforced concrete structures. Estimating reinforcing bar (rebar) parameters from GPR radargrams remains challenging due to the strong correlation among rebar-related parameters when producing the reflection signature. Additionally, the unknown rebar orientation affects GPR detection capabilities and adds further difficulties in rebar parameter estimation, which has not yet been addressed in the existing literature. To tackle these issues, we present a neural network structure, called Multi-Polarimetric Aggregation and Parameter Decorrelation Neural Network (MAPD-Net), to automatically derive multiple rebar and concrete parameters from GPR radargrams. These parameters include concrete moisture content (
mc
), rebar cover depth (
d
), radius (
r
), and orientation (φ). Given multi-polarization radargrams as inputs, the MAPD-Net extracts informative features from each polarization radargram, weakens parameter correlation, and performs estimation of each parameter. Numerical results demonstrate that the MAPD-Net achieves high estimation accuracy in estimating these parameters. The mean absolute errors of
mc
,
d
,
r
, and φ in the 600 testing data are 0.06%, 0.8 mm, 0.4 mm, and 1.0°, respectively. The average absolute percentage errors of them are 2.2%, 1.1%, 3.8%, 4.4%, respectively.