ML-Enhanced Ground Penetrating Radar for Soil Structure Analysis

Abstract

Ground Penetrating Radar (GPR) is a widely used geophysical tool for subsurface exploration, providing high-resolution imaging of underground structures. However, interpreting raw GPR B-scan data is challenging due to noise, soil heterogeneity, and attenuation effects caused by varying moisture and permittivity. This study presents a machine learning-enhanced GPR system designed to automatically process B-scan data and extract meaningful subsurface features. The system integrates advanced preprocessing techniques, including depth-dependent gain correction, denoising via Gaussian filtering, and intelligent thresholding, to enhance signal quality. The proposed framework combines soil parameters, such as moisture content, permittivity, and soil type, with user-configurable machine learning parameters to dynamically adapt the analysis. Users interact with the system via a custom GUI built with Custom Tkinter, enabling seamless data upload, real-time analysis, visualization of raw and processed B-scans, and interactive exploration of results. The system allows synthetic dataset generation for testing, and robust error handling ensures reliable operation even with imperfect datasets.The ML component employs adaptive thresholding and deep feature recovery techniques to isolate meaningful reflections from noise and clutter, simulating intelligent interpretation of subsurface structures. Corrected data is visualized alongside the raw GPR B-scan for comparison, providing clear insights into soil stratigraphy and buried object detection potential. Additionally, the system calculates propagation velocity based on permittivity, supporting depth migration and accurate layer estimation. Experimental evaluation demonstrates the system’s effectiveness in revealing hidden structures and compensating for soil-specific attenuation effects. The modular design allows for future integration of advanced models or real-time field deployments. Overall, this ML-enhanced GPR system bridges the gap between conventional radar interpretation and intelligent data-driven analysis, offering a practical tool for geotechnical engineering, archaeology, and environmental studies.