Abstract
This study introduces a practical method for diagnosing subsurface structures which focuses on roads and uses ground-penetrating radar (GPR) and deep learning.The study proposes an implementation to interpret radargrams captured by the OKO-2 system, which records signals reflected from underground layers.Two interpretation methods are used, i.e., one based on physics formulas, the other on mathematical modeling.Before applying neural networks, the data undergo analytical processing using the GeoScan32 software, which filters noise, enhances signals, and identifies layer boundaries.This analytical foundation is crucial for effectively training deep learning models to interpret complex subsurface data.