Deep learning based method for 3D reconstruction of underground pipes in 3D GPR C-scan data
发布时间：2025-01-07 点击次数：
DOI码：10.1016/j.tust.2024.105819
所属单位：School of Remote Sensing and Information Engineering, Wuhan University, China
发表刊物：TUNNELLING AND UNDERGROUND SPACE TECHNOLOGY
关键字：Ground penetrating radar,C -scan data,Underground pipe reconstruction,Semantic segmentation model,Deep learning
摘要：Many urban underground locations suffer from incomplete or inaccurate records. Nevertheless, a comprehensive understanding of the subsurface conditions is crucial, particularly for pipe-related projects. Ground Penetrating Radar (GPR) serves as a widely-used nondestructive technique for detecting and locating underground objects. The efficient extraction of information from GPR data has been a research objective, especially in areas related to 3D GPR, which lacks sufficient investigation. In this paper, we propose a comprehensive workflow for constructing 3D models of underground pipes. Our approach differs from common classification tasks as it harnesses the 3D information within the GPR data and optimizes traditional 3D GPR processing methods. Additionally, for the semantic segmentation of pipes within 3D GPR data, we have made a series of attempts to improve it based on the 3D Unet network. Experimental results confirm the effectiveness of our method in reconstructing underground pipes using the 3D GPR data.
合写作者：Huang,Ai,Yu, Qingwu,Hu, Ju,Zhang, Yibo,Zhou,Zhao, Yuchun, Mingyao, Fei, Pengcheng
论文类型：期刊论文
学科门类：工学
文献类型：J
卷号：150
ISSN号：0886-7798
是否译文：否
CN号：EI:20242116138451,WOS:001246517100001,Scopus:2-s2.0-85193809066
发表时间：2024-08-01
