Optimization of data sampling and image reconstruction by GPR

This paper focuses on 3-dimensional (3D) image reconstruction by ground penetrating radar (GPR) data. Conventionally, we acquired a GPR gridded dataset with a fine interval, which satisfies the Nyquist spatial sampling criterion for an antenna. However, it takes long time for data acquisition. In this study, we tried two different approaches to reconstruct the image with sparse data that violated the Nyquist spatial sampling criterion: A non-gridded 3D migration method and a new interpolation method based on Projection onto convex sets (POCS) and frequency-wave number (f-k) filtering. Both methods are demonstrated with sand pit experiment datasets and a field experiment data that is acquired by our 3DGPR system. The results shows that both the non-gridded 3D migration method and the interpolation method can reconstruct the main target (a metal pipe at 0.8 m depth) well with the average spatial interval that equals to half wave length. But the non-gridded migration results (especially in shallow depth) suffer from the migration artifacts. The migrated result after interpolation is also demonstrated, and the migration artifacts can be reduced. These results indicate that it is possible to reduce the data density.