Cost-effective image recognition of water leakage in metro tunnels using self-supervised learning

Authors

Yining Gu
Qing Ai
Zhengwen Xu
Lianbi Yao
Hui Wang
Xingchun Huang
Yong Yuan

Document Type

Journal Article

Publication Date

2024

Subject Area

mode - subway/metro, infrastructure - maintainance, place - urban, planning - methods, planning - safety/accidents

Keywords

metro tunnels, self-supervised learning (SSL), image recognition

Abstract

Supervised learning (SL) methods have achieved good performance in image recognition of water leakage in tunnels, but they require a large amount of labeled data. This paper adopts a self-supervised learning (SSL) approach to realize cost-effective image recognition of water leakage in metro tunnels to reduce the necessary labeling effort. The Bootstrap Your Own Latent (BYOL) for contrastive learning, a framework of SSL method, is used to classify tunnel images with only 10% labeled data, and the model is transferred to two other tunnel datasets to analyze its ability for practical application. The results show that the proposed SSL method outperforms SL method with the most cost-effective data labeling proportion of approximately 10%, and the transferability of the SSL model is also significantly better than that of the SL model. This method is promising for reducing the labeling effort and improving the transferability of tunnel image recognition.

Rights

Permission to publish the abstract has been given by Elsevier, copyright remains with them.

Comments

Automation in Construction Home Page:

http://www.sciencedirect.com/science/journal/09265805

Recommended Citation

Gu, Y., Ai, Q., Xu, Z., Yao, L., Wang, H., Huang, X., & Yuan, Y. (2024). Cost-effective image recognition of water leakage in metro tunnels using self-supervised learning. Automation in Construction, 167, 105678.