Scholarlink

<p>Deep learning methodologies can enable computers to replace humans in identifying underground diseases within Ground Penetrating Radar (GPR) images. This method has the potential to significantly enhance the efficiency of subsurface anomaly detection. In order to improve the discriminative capability and precision of GPR imagery interpretation for subsurface voids and pipelines, this study used a group of GPR images acquired through field measurements. The training dataset was enlarged through software-based forward simulation and augmented by CycleGAN. After that, a YOLO v8 model was trained to achieve automated classification of the GPR images. The results show that: (1) The YOLO v8 model demonstrates proficiency in localizing and distinguishing three subterranean objects: underground voids, metal pipelines, and concrete pipelines. (2) The utilization of forward modeling simulation contributes to an enhancement in training precision and recall of the model. However, the proportion of data augmentation should not be too high. (3) The transformation between the measured and forward simulation images can significantly improve recognition accuracy through data augmentation. The automatic recognition method proposed in this study can effectively improve the discrimination speed and accuracy of GPR images and positively affect the rapid detection of underground space diseases and structures.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Research on Automatic Recognition Technology of Underground Voids and Pipelines in Ground Penetrating Radar Images

  • 期刊论文

  • 发表日期：

    2025年5月14日

查看全文
  • Yunxi Han,
  • Yupeng Shen,
  • Jiayi Xu,
  • Jamlick Mwangi Kariuki,
  • Zhiqiang Li

摘要

Deep learning methodologies can enable computers to replace humans in identifying underground diseases within Ground Penetrating Radar (GPR) images. This method has the potential to significantly enhance the efficiency of subsurface anomaly detection. In order to improve the discriminative capability and precision of GPR imagery interpretation for subsurface voids and pipelines, this study used a group of GPR images acquired through field measurements. The training dataset was enlarged through software-based forward simulation and augmented by CycleGAN. After that, a YOLO v8 model was trained to achieve automated classification of the GPR images. The results show that: (1) The YOLO v8 model demonstrates proficiency in localizing and distinguishing three subterranean objects: underground voids, metal pipelines, and concrete pipelines. (2) The utilization of forward modeling simulation contributes to an enhancement in training precision and recall of the model. However, the proportion of data augmentation should not be too high. (3) The transformation between the measured and forward simulation images can significantly improve recognition accuracy through data augmentation. The automatic recognition method proposed in this study can effectively improve the discrimination speed and accuracy of GPR images and positively affect the rapid detection of underground space diseases and structures.