Optimized GPR Migration Imaging for Accurate Underground Pipeline Detection via a Dual-Stage Unified Deep Learning Framework

Ground Penetrating Radar (GPR) is a crucial geophysical tool for near-surface detection. Migration imaging can delineate structural information of underground targets. However, limited by the inherent inadequate resolution of the migration methods and the effects of crosstalk artifacts, it is challenging to clearly differentiate adjacent objects. This paper introduces a dual-stage unified deep learning framework centered on an Edge-Enhanced Feature Distillation Network (EFDN) designed for the accurate detection of underground pipelines. Utilizing edge detection filters and attention mechanisms, the framework not only refines feature details during the imaging process but also effectively suppresses interference signals. Through a series of numerical simulation, laboratory experiment, and field test, this paper confirms the effectiveness of the proposed unified deep learning framework. Applied to resolution enhancement and reflectivity imaging-guided artifact suppression, the framework enhances the quality of migration imaging in practical applications, thereby demonstrating its potential for the accurate detection of underground pipelines.