An efficiency-perception joint-driven lightweight network for subterranean pipeline defect detection
Measurement Science and Technology · 07.08.2026 · 37 (31) · 315403
DOI: 10.1088/1361-6501/ae8ded ↗Kurz gefasst
Fachartikel in der Zeitschrift Measurement Science and Technology, online erschienen am 04.08.2026. Verfasst von Sunke Zhu und 4 weiteren Autor:innen. Die Arbeit nennt 43 Literaturangaben. Gefördert von: Chongqing Municipal Key Research and Development Program of China.
Abstract (Originalsprache)
Detecting internal defects in subterranean pipelines is crucial for ensuring the reliable operation and structural integrity of underground engineering infrastructure. However, field deployment remains challenging because defects often have low-contrast boundaries and large variations in scale, while embedded inspection platforms impose strict computation, memory-access, and latency constraints. Existing visual inspection algorithms impose high computational demands, making it challenging to balance detection performance with hardware resource consumption on embedded devices. To address these issue, this study proposed EP-YOLOv8, an optimized efficiency-perception joint-driven lightweight algorithm based on YOLOv8 and tailored for low-power edge deployment. Firstly, we propose an efficiency-perception joint-driven backbone combining partial convolution (PConv) and squeeze-and-excitation version 2, which reduces redundant memory access while adaptively recalibrating channel-wise features to highlight low-contrast defect boundaries. Secondly, we construct a memory access cost-aligned bidirectional feature pyramid network neck composed of grouped spatial convolution and C2f-PConv modules. As the experimental results demonstrate, compared to the original model, the EP-YOLOv8 algorithm achieves an 4.0% improvement in mean average precision (mAP50). Crucially, this performance is achieved with a drastic reduction in computational resources: the number of parameters is reduced by 59.8% and the computational complexity (GFLOPs) is reduced by 61.7%. The model was also deployed on a Raspberry Pi 4B-based inspection platform, supporting its potential for on-device screening and localization of visible pipeline defects under resource constraints.
Förderung
- Chongqing Municipal Key Research and Development Program of China
Zitieren
APA
Zhu, S., Zeng, D., Tan, H., Ni, Y., & Xiong, Y. (2026). An efficiency-perception joint-driven lightweight network for subterranean pipeline defect detection. Measurement Science and Technology, 37(31), 315403. https://doi.org/10.1088/1361-6501/ae8ded
Vancouver
Zhu S, Zeng D, Tan H, Ni Y, Xiong Y. An efficiency-perception joint-driven lightweight network for subterranean pipeline defect detection. Measurement Science and Technology. 2026;37(31):315403. doi:10.1088/1361-6501/ae8ded
BibTeX
@article{Zhu2026,
author = {Zhu, Sunke and Zeng, Daqiao and Tan, Hongwei and Ni, Yingqian and Xiong, Yikun},
title = {An efficiency-perception joint-driven lightweight network for subterranean pipeline defect detection},
journal = {Measurement Science and Technology},
year = {2026},
volume = {37},
number = {31},
pages = {315403},
doi = {10.1088/1361-6501/ae8ded}
}
EUVIMED