Enhancing Interpretability in Deep Learning-Based Inversion of 2-D Ground Penetrating Radar Data: An Explainable AI (XAI) Strategy

Recent advancements in deep learning (DL) have demonstrated potential for interpreting ground-penetrating radar (GPR) data, which is crucial for near surface geophysical investigations. However, the complexity of DL models and the challenge of interpreting their decision-making processes remain significant obstacles. This study addresses these challenges by applying explainable AI (XAI) techniques—specifically, local interpretable model-agnostic explanations (LIMEs) and gradient-weighted class activation mapping (Grad-CAM)—to elucidate the DL-based inversion process for 2-D GPR data. Our novel approach marks the first application of these XAI techniques in the context of GPR data analysis for subsurface utility mapping, revealing critical features and hierarchical feature extraction processes that drive the model’s predictions. By offering detailed insights into the model’s internal operations, this research not only enhances the interpretability of DL models in geophysical applications but also establishes a new standard for incorporating XAI in subsurface utility detection, paving the way for more accurate, reliable, and understandable DL applications in geophysical studies.