U-Net++ GPR field Data

This dataset contains field-measured ground-penetrating radar (GPR) B-scan profiles acquired at a shallow karst test site in Guilin, China. Data were collected using a MALA ProEx GPR system equipped with a shielded 250 MHz Rough Terrain Antenna (RTA) in single-offset, common-offset configuration. Trace positioning was achieved using an RTK-GNSS rover with ±2 cm horizontal accuracy, and surveys were conducted at 1 cm spatial intervals using a calibrated survey wheel. The dataset comprises four GPR profiles, each containing 128 traces and 512 time samples over a time window of approximately 60 ns sampled at 0.12 ns, corresponding to an investigation depth of approximately 2–4 m. Standard preprocessing was applied to all profiles, including bandpass filtering, gain correction, and background subtraction. This dataset was used to evaluate deep learning-based GPR data reconstruction using a multi-scale U-Net++ framework, as described in the associated publication. It is made openly available to support reproducibility and to facilitate further research in near-surface geophysical data reconstruction and karst subsurface characterization.