Shape-Aware Topological Representation for Pipeline Hyperbola Detection in GPR Data

Ground penetrating radar (GPR) is a widely used nondestructive testing (NDT) technique for subsurface exploration, particularly in infrastructure inspection. However, traditional interpretation methods often struggle with noise sensitivity and limited structural awareness. We propose a novel framework that integrates shape-aware topological features, derived from B-scan GPR images using topological data analysis (TDA), with the object detection capabilities of a YOLOv5-based deep neural network (DNN). This topological representation improves geometric salience, enhancing detection and localization of underground utilities, especially pipelines. To mitigate the scarcity of annotated realworld data, a sim-to-real (Sim2Real or S2R) strategy is employed. Synthetic datasets are generated to capture both diverse subsurface conditions and the essential hyperbolic reflection patterns of pipelines, enabling more effective knowledge transfer to real-world scenarios. Experimental results show consistent improvements in mean average precision (mAP), highlighting the robustness and effectiveness of the proposed method. This work demonstrates the potential of TDA-enhanced deep learning for reliable subsurface object detection with broad implications in urban planning, safety inspection, and infrastructure management.