EMDAInvNet: A Deep Learning Framework for 3-D Permittivity Inversion of Underground Pipelines
IEEE Sensors Journal · 01.06.2026 · 26 (11) · 17527-17540
DOI: 10.1109/jsen.2026.3687430 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Sensors Journal, erschienen am 01.06.2026. Verfasst von Lingcen Kong und 9 weiteren Autor:innen. Die Arbeit nennt 45 Literaturangaben. Gefördert von: Natural Science Foundation of Shandong Province.
Abstract (Originalsprache)
The spatial positioning and structural information of underground pipelines are crucial for urban construction and operation. Traditional methods based on hyperbolic curve fitting and time-distance inversion rely heavily on local geometric features and often fail in complex media, strong clutter, and multitarget overlapping scenarios. To specifically address these limitations, this article proposes an end-toend inversion network, enhanced multiscale dilated-attention 3-D inversion network (EMDAInvNet), for ground penetrating radar (GPR) 3-D data. The network takes GPR volumetric data as input and directly outputs the geometric information of underground pipelines in 3-D space. The network follows an encoder–decoder architecture, where at each encoding and decoding stage, an enhanced multiscale fusion attention (EMSFA) and DilatedBlock module are cascaded. Specifically, the EMSFA module utilizes channel and spatial attention mechanisms to adaptively suppress background clutter, while the DilatedBlock captures global topological context to distinguish adjacent targets in overlapping scenarios. The model enables direct visualization of the pipeline’s shape, orientation, and relative position. Numerical data and experimental results show that the proposed method improves the accuracy and robustness of GPR-based 3-D inversion, offering a feasible and efficient technological approach for nondestructive testing of complex urban underground spaces.
Förderung
- Natural Science Foundation of Shandong Province
Zitieren
APA
Kong, L., Zhang, J., Lian, F., Liu, Z., Hou, Z., Liu, J., Lu, J., Li, Y., Yang, Z., & Lu, Y. (2026). EMDAInvNet: A Deep Learning Framework for 3-D Permittivity Inversion of Underground Pipelines. IEEE Sensors Journal, 26(11), 17527-17540. https://doi.org/10.1109/jsen.2026.3687430
Vancouver
Kong L, Zhang J, Lian F, Liu Z, Hou Z, Liu J, et al. EMDAInvNet: A Deep Learning Framework for 3-D Permittivity Inversion of Underground Pipelines. IEEE Sensors Journal. 2026;26(11):17527-17540. doi:10.1109/jsen.2026.3687430
BibTeX
@article{Kong2026,
author = {Kong, Lingcen and Zhang, Jiong and Lian, Feng and Liu, Zhi and Hou, Zhao and Liu, Jinlong and Lu, Jiaqi and Li, Yanan and Yang, Zhen and Lu, Yang},
title = {EMDAInvNet: A Deep Learning Framework for 3-D Permittivity Inversion of Underground Pipelines},
journal = {IEEE Sensors Journal},
year = {2026},
volume = {26},
number = {11},
pages = {17527--17540},
doi = {10.1109/jsen.2026.3687430}
}
EUVIMED