A Multi-Task Framework for Interpreting Pavement Subsurface Distress From GPR Data With Multimodal Vision-Language Model (opens in new tab)

In recent years, deep learning (DL) has become an effective tool for analyzing ground penetrating radar (GPR) data. It enables the rapid identification of pavement subsurface distress. However, traditional single-task DL models often fail to provide comprehensive and effective interpretation in real engineering scenarios. To address this problem, this article proposes a multi-task interpretation framework based on the multimodal Vision-Language model PaliGemma, along with an efficient annotat...