Reconstruction of Missing Ground-Penetrating Radar Traces Using Simplified U-Net

Ground-penetrating radar is an effective tool for the exploration of subsurface and obscured objects by exploiting the electromagnetic wave characteristics. However, some problems in the measurement process produce missing traces. The resulting incomplete data create difficulties in visualization, interpretation, and analysis. This article proposes a lightweight deep learning method to reconstruct the missing traces. The model is a simplification of a well-known convolutional neural network model, U-Net, which performs well in this context. We use a multi-and-wide kernel size, shallow convolutional layers, a reduced encoding-decoding step, and a minimized feature map size to minimize the model while maintaining high performance. In this study, we focus on the case of inspection of reinforced concrete, where the examined structure consists of a series of rebar and/or void cracks. The numerical and experimental field data results indicate that the method can reconstruct the missing traces with high accuracy. The proposed method also shows comparable performance with U-Net, although the model size is more than 45 times smaller, and its computation speed is more than three times faster. Thus, the proposed method is promising for use in a wide range of applications and devices.

Automated summary

This article proposes a lightweight deep learning method for reconstructing missing traces in ground-penetrating radar data. The method simplifies the model structure while maintaining high performance, enabling accurate reconstruction with improved speed and computation. The method shows promising potential for applications such as reinforced concrete inspection.