Automated Subsurface Utility Detection with Ground Penetrating Radar

Ground-penetrating radar (GPR) remains a primary tool for subsurface utility detection, and its efficacy is often compromised by complex data patterns and environmental perturbations. Traditional GPR analysis can be inadequate, necessitating advanced methodologies for precise interpretation. In this study, we introduce an integration of Convolutional Neural Networks (CNN) with GPR to improve the detection and delineation of subsurface utilities systematically. The CNN model is trained on diverse GPR datasets, taking into account the various soil conditions and utility materials that influence radar wave reflections and refractions. To ensure the robustness and applicability of our CNN-enhanced GPR analysis, we established a comprehensive test bed at the Kennesaw State University Field Station. This test bed, measuring 15 feet in length and 5 feet in depth, is equipped with controlled soil densities and known buried objects. The controlled variability within the test bed provides a realistic platform to challenge and refine our deep learning approach, bridging the gap between laboratory experimentation and field application. Initial results underscore satisfactory results in subsurface utility detection precision, demonstrating the significant potential of our CNN-GPR integration. With this approach, challenges associated with GPR data variability can be effectively addressed, ensuring more reliable and accurate subsurface utility mapping in civil engineering projects.