Exploring Beneath the Surface: Mapping the Underground with AI and Radar
Disciplines
Civil Engineering
Abstract (300 words maximum)
Ground-Penetrating Radar (GPR) is a vital technology for non-invasive subsurface exploration, enabling the identification of buried objects and structural anomalies. A key challenge in interpreting GPR data is the reliable recognition of hyperbolic signatures within noisy B-scans. In this work, we present a deep learning-based framework for automated hyperbola detection that directly learns from real GPR data, eliminating the need for manual feature engineering. Leveraging a carefully curated dataset of annotated radar images, our convolutional neural network model extracts and exploits high-level features to distinguish hyperbolic reflections from clutter and other background interferences. Experimental evaluations demonstrate that our approach achieves superior accuracy and robustness compared to traditional methods, reducing both false positives and missed detections. This advancement not only accelerates GPR data processing but also improves the reliability and scalability of subsurface mapping. By harnessing the power of AI and radar, we provide an automated solution that can be easily integrated into existing workflows, paving the way for more efficient geological surveys, infrastructure assessments, and archaeological explorations.
Academic department under which the project should be listed
SPCEET - Civil and Environmental Engineering
Primary Investigator (PI) Name
Da Hu
Exploring Beneath the Surface: Mapping the Underground with AI and Radar
Ground-Penetrating Radar (GPR) is a vital technology for non-invasive subsurface exploration, enabling the identification of buried objects and structural anomalies. A key challenge in interpreting GPR data is the reliable recognition of hyperbolic signatures within noisy B-scans. In this work, we present a deep learning-based framework for automated hyperbola detection that directly learns from real GPR data, eliminating the need for manual feature engineering. Leveraging a carefully curated dataset of annotated radar images, our convolutional neural network model extracts and exploits high-level features to distinguish hyperbolic reflections from clutter and other background interferences. Experimental evaluations demonstrate that our approach achieves superior accuracy and robustness compared to traditional methods, reducing both false positives and missed detections. This advancement not only accelerates GPR data processing but also improves the reliability and scalability of subsurface mapping. By harnessing the power of AI and radar, we provide an automated solution that can be easily integrated into existing workflows, paving the way for more efficient geological surveys, infrastructure assessments, and archaeological explorations.