Answers are grounded in this paper and cite their sources.
Detecting subsurface voids is crucial for sustainable road maintenance to ensure traffic safety. Ground Penetrating Radar (GPR) is pivotal for the subsurface assessments, yet traditionally relies on manual image analysis, limiting inspection frequency and risk identification. This study introduces a novel 3D unsupervised void detection method including 3 stages: first, distilling a pretrained video classification model to adapt it for 3D grayscale data. Second, training a student model and an autoencoder on an unlabelled GPR dataset using the distilled teacher model to align outputs of the student model and auto encoder. In the third stage, discrepancies between the student model and autoencoder outputs versus the teacher model highlight potential voids. Tests on real-world GPR data demonstrate the method's efficacy in accurately localizing voids, showing the improvements of both classification and localization accuracy comparing with baseline methods.
No takes yet. Share an insight, caveat, or question.
Chen et al. (2025) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: