Attention-Enhanced Multi-Task CNN for Subway Tunnel Lining Crack Segmentation and Defect Grading with Lightweight Deployment
Abstract
This study proposes a multi-task convolutional neural network (CNN) with a ResNet-34 backbone, CBAM attention modules, and a multi-scale fusion head for crack segmentation and defect grading in subway tunnel linings. The model integrates shared feature extraction with two task-specific heads, enabling precise crack edge segmentation and severity estimation in a unified framework. Experiments on a dataset of 12,000 RGB and multispectral images (8,400/2,400/1,200 for training/validation/testing) showed that the proposed model achieved mIoU = 91.2% ± 1.0, Dice/F1 = 93.0% ± 0.8, and mAP@0.5 = 90.7% ± 0.9 on the test set. Recognition accuracy reached 94.3%, exceeding a rule-based method (78.9%) and four deep models—U-Net, DeepLabV3+, PSPNet, and Faster R-CNN (≈88%). Evaluation replaced 'recognition accuracy' with segmentation/detection metrics: pixel-F1, mIoU, boundary F-score (BSDS), AP50-95 for instance cracks, and macro/micro-F1 for grade prediction. Per-crack type and per-grade metrics, ROC, calibration (ECE/Brier), confusion matrices, and bootstrap CIs were also reported.Average inference latency was 1.8 ± 0.2 s, with a response delay of 0.9 ± 0.1 s and an interruption rate of 2.5%, while CPU usage remained below 30% on an Intel i5 platform. Even with 10% noise, accuracy stayed at 92.1%, demonstrating strong robustness. These results confirm that the proposed framework combines accuracy, speed, and stability, supporting real-time deployment for tunnel-lining crack inspection.References
Huang H, Zhao S, Zhang D, et al. Deep learning-based instance segmentation of cracks from shield tunnel lining images[J]. Structure and Infrastructure Engineering, 2020,16(14):1826–1840.https://doi.org/10.1080/15732479.2020.1838559.
Zhao S, Zhang D, Xue Y, Zhou M, Huang H. A deep learning-based approach for refined crack evaluation from shield tunnel lining images[J]. Automation in Construction,2021,132:103934.https://doi.org/10.1016/j.autcon.2021.103934.
Zhou Z, Zheng Y, Zhang J, Yang H. Fast detection algorithm for cracks on tunnel linings based on deep semantic segmentation[J]. Frontiers of Structural and CivilEngineering,2023,17:732–744.https://doi.org/10.1007/s11709-023-0965-y.
Yang K, et al. Deep learning-based YOLO for crack segmentation and measurement in metro tunnels[J]. Automation in Construction,2024.https://doi.org/10.1016/j.autcon.2024.105818.
Huang H W , Li Q T , Zhang D M .Deep learning based image recognition for crack and leakage defects of metro shield tunnel - ScienceDirect[J].Tunnelling and Underground Space Technology, 2018, 77:166-176.https://doi.org/10.1016/j.tust.2018.04.002.
Mei C, Wen Y. Subway tunnel crack identification based on YOLOv5[J]. Frontiers in Computing and Intelligent Systems,2024,8(1):122129.https://doi.org/10.54097/7gw4nw71.
Sun W, Liu X, Lei Z. A tunnel crack segmentation and recognition algorithm using SPGD-and-generative adversarial network fusion[J]. Sensors, 2025, 25(8): 2381. https://doi.org/10.3390/s25082381.
Wang L, Tang C. Effective small crack detection based on tunnel crack characteristics and an anchorfree convolutional neural network[J]. Scientific Reports, 2024, 14: 10355. https://doi.org/10.1038/s41598-024-60454-3.
Lee K, Lee S, Kim HY. Deep learning-based defect detection framework for ultra high resolution images of tunnels[J]. Sustainability, 2023, 15(2): 1292. https://doi.org/10.3390/su15021292.
Feng Y, et al. Automatic classification and segmentation of tunnel lining cracks using two-step deep learning based method[J]. preprint(ArXiv),2025.https://doi.org/10.48550/arXiv.2507.14010.
Nyathi M A , Bai J , Wilson I D .Deep Learning for Concrete Crack Detection and Measurement[J].Metrology, 2024, 4(1).https://doi.org/10.3390/metrology4010005.
Krishnan SSR, et al. Comparative analysis of deep learning models for crack detection in concrete[J]. Scientific Reports,2025.https://doi.org/10.1038/s41598-025-85983-3.
Xu G , Yue Q , Liu X .Deep learning algorithm for real-time automatic crack detection, segmentation, qualification[J].Engineering Applications ofArtificialIntelligence,2023,126(PartC):22.https://doi.org/10.1016/j.engappai.2023.107085.
Zhou Z , Zhang J , Gong C ,et al.Automatic tunnel lining crack detection via deep learning with generative adversarial network-based data augmentation[J].Underground Space, 2023,9:140-154.https://doi.org/10.1016/j.undsp.2022.07.003.
Dang L M, Wang H, Li Y, Park Y, Oh C, Nguyen T N, Moon H. Automatic tunnel lining crack evaluation and measurement using deep learning[J]. Tunnelling and Underground Space Technology, 2022, 124: 104472.https://doi.org/10.1016/j.tust.2022.104472.
Li L , Yang Y , Bian M ,et al.SnakeConv and SFC boosting precise segmentation on the crack of tunnel lining surface: based on DeepLabV3+with improved Swin transformer V2[J].Measurement Science &Technology,2025(2):36.https://doi.org/10.1088/1361-6501/ada2b6.
Dongming Zhang, et al. UnrollingNet: attention-based deep learning approach for tunnel point clouds segmentation[J]. Automation in Construction, 2022. https://doi.org/10.1016/j.autcon.2022.104456.
Ji A, Zhang L, Fan H, et al. Dual attention-based deep learning network for multi-class object semantic segmentation of tunnel point clouds[J]. Automation in Construction,2023,145:105131.https://doi.org/10.1016/j.autcon.2023.105131.
Xu L, Wang Y, Dong A, et al. Image-based intelligent detection of typical defects of complex subway tunnel surface[J]. Tunnelling and Underground Space Technology,2023,140:105266.https://doi.org/10.1016/j.tust.2023.105266.
Wu J, Zhang X. Tunnel crack detection method and crack image processing algorithm based on improved Retinex and deep learning[J]. Sensors, 2023, 23(22): 9140. https://doi.org/10.3390/s23229140.
DOI:
https://doi.org/10.31449/inf.v49i4.11278Downloads
Published
How to Cite
Issue
Section
License
Authors retain copyright in their work. By submitting to and publishing with Informatica, authors grant the publisher (Slovene Society Informatika) the non-exclusive right to publish, reproduce, and distribute the article and to identify itself as the original publisher.
All articles are published under the Creative Commons Attribution license CC BY 3.0. Under this license, others may share and adapt the work for any purpose, provided appropriate credit is given and changes (if any) are indicated.
Authors may deposit and share the submitted version, accepted manuscript, and published version, provided the original publication in Informatica is properly cited.
