Code and Example Data Supporting Deep Learning-Based Dielectric Permittivity Inversion of RIMFAX GPR Data

The “Data” folder contains two example data files, Ep_sample and RIMFAX_Shallow_sample, each consisting of the first 1,000 rows extracted from the training dataset and provided for demonstration and reproducibility purposes only. These files represent a small subset of the full dataset used in the study.

RIMFAX_Shallow_sample stores one-dimensional corrected FDTD-simulated ground-penetrating radar (GPR) signals with added random Gaussian noise, which serve as the network inputs. The corresponding target outputs are provided in Ep_sample, which contains the associated one-dimensional dielectric permittivity profiles.

The “Code” folder contains four files. GPRNet implements the detailed architecture of the proposed deep learning model. Script 1 (“Split data set into training, testing and validation”) partitions the data into training, validation, and test sets. Script 2 (“Train GPR data and velocity_PyTorch_Shallow”) is used to train the model, and Script 3 (“Show the training process_Shallow”) records and visualizes the training process.

The code has been developed and tested using Python 3.10.16 (Anaconda distribution) and PyTorch 2.8.0.dev20250408+cu128 on an NVIDIA GeForce RTX 5080 GPU. The provided example data allow users to execute the scripts and reproduce the workflow without access to the full dataset. When running the code on other systems, users only need to modify the file paths to match their local directory structure.