An Image Deconvolution Method for Ground Penetrating Radar Based on Generative Adversarial Network

Ground penetrating radar (GPR) represents an established non-destructive detection technology. Enhancing resolution of low-frequency GPR data of a broad detection enables acquisition of high-quality GPR images, which is of great significance for interpreting and processing GPR data. Nevertheless, historical research methods have encountered challenges including decreased signal-to-noise ratio, instability in effects, computational complexity, etc. This paper introduces a generative adversarial network-based approach to enhance low-frequency GPR image resolution. The proposed technique effectively disentangles overlapping echoes and compresses hyperbolas to yield high-resolution images. In the field experiment, the proposed method is also successfully applied to the data of different frequency other than the training data, which proves the effectiveness and generalization of the proposed method.