Open Access

Reza Aliabadi, Mathias Kromer, Marlene Harter

• Prior knowledge of subsurface layers is crucial in many applications, particularly infrastructural projects. Ground Penetrating Radar (GPR) is a preferred measurement technique for investigating subsurface layers due to its non-destructive nature when probing the ground. However, interpreting GPR data is not trivial and requires the experience of the user. In this paper, the possibility of usingPrior knowledge of subsurface layers is crucial in many applications, particularly infrastructural projects. Ground Penetrating Radar (GPR) is a preferred measurement technique for investigating subsurface layers due to its non-destructive nature when probing the ground. However, interpreting GPR data is not trivial and requires the experience of the user. In this paper, the possibility of using one-dimensional Convolutional Neural Networks (1D CNNs) is investigated to estimate the thickness of up to two different subsurface layers with different electromagnetic characteristics. For this purpose, the network is trained on a synthetically generated stepped-frequency GPR A-scan dataset. The predicted thickness results using the trained 1D CNN showed good accuracy with a relative error of 2.6 % and 9.2 % for the first and second layers, respectively.…