To address the challenges in detecting surface defects in subway tunnels, such as blurred boundaries, multi-scale variations, slender shapes, and class imbalance, an improved SegFormer-ESF model was proposed. First, a parallel edge information enhancement branch was designed. Fixed Sobel operators and depthwise separable convolutions were used to extract and enhance gradient features. These edge features were adaptively fused with backbone features through a feature fusion module, which improved perception and localization of defect boundaries. Second, a directional strip convolutional attention module was embedded in the encoder. It employed 1×K and K×1 convolutions to model geometric structure and directional context of linear defects, such as cracks and leakage, thereby enhancing representation of slender objects. Then, a lightweight multi-scale pyramid feature fusion module was introduced to reconstruct the decoder, improving multi-scale feature integration while reducing model complexity. Finally, to mitigate class imbalance caused by low pixel proportion of defects, a combined Focal-Dice loss function was proposed, which jointly optimized hard example mining and overall region recognition. Experimental results show that SegFormer-ESF achieves 83.53%, 92.46%, and 97.61% on mIoU, mPA, and Acc, respectively, improving by 3.72%, 3.51%, and 0.46% over SegFormer-B0. In addition, the model has 3.620 M parameters and only 3.02 G computational cost, reduced by 2.56% and 55.46% compared with the baseline, respectively, achieving lightweight while improving segmentation accuracy. The FPS reaches 88.53 f/s, which can meet the demand for real-time detection, providing a feasible solution for intelligent identification and mobile deployment of tunnel defects.
WU Xiao-chun, JIANG Kai.
Method for Detecting Surface Defects in Subway Tunnels by Integrating Edge Features and Directional Perception[J].
Science Technology and Engineering, 2026, 26(30): 13145-13157 DOI:10.12404/j.issn.1671-1815.2507627
地铁隧道内部环境复杂,其背景中的光影变化、衬砌接缝与支撑构件等干扰因素与病害区域的视觉特征相似,增加了病害与背景的区分难度。此外,裂缝、剥落等病害本身形态细微,导致现有语义分割模型在精确识别病害区域时面临挑战。边缘信息能有效刻画此类细微的形态差异,有助于强化局部特征判别力,进而提升病害边界的分割精度。然而,传统语义分割方法往往难以有效协调低级边缘定位与高级语义识别之间的关系。为解决上述问题,提出一种边缘信息增强(edge information enhancement, EIE)分支,并将其引入SegFormer模型中,旨在通过建模图像边缘结构,进而与主干网络的语义特征相融合,以增强模型在复杂隧道环境下对病害边界与细节特征的感知能力。
China Association of Metros. Essential interpretation of 2024 Statistical and Analysis Report on Urban Rail Transit[J]. China Metros, 2025(4): 22-25.
[3]
ZhouZ, LiS, YanL, et al. Intelligent recognition of tunnel lining defects based on deep learning: methods, challenges and prospects[J]. Engineering Failure Analysis, 2025, 170: 109332.
[4]
ZhangW, TianC, YangJ, etal. Vision and deep learning-based tunnel surface defect detection: technological evolution, challenges, and prospects[J]. Measurement, 2025, 230: 118853.
[5]
TanL, HuX, TangT, et al. A lightweight metro tunnel water leakage identification algorithm via machinevision[J]. Engineering Failure Analysis, 2023, 150: 107327.
LiuXian, SunQihao. Case analysis on progressive collapse of shield tunnel linings[J]. Hazard Control in Tunnelling and Underground Engineering, 2020, 2(2): 21-30.
WuXiaochun, LiLuyu. Tunnel bolt rust detection combining local enhancement and improvement of YOLOv8[J]. Computer Engineering and Applications, 2025, 61(19): 249-259.
LiaoWeizhang, HuangShuchen, YuanWanying, et al. Intelligent detection algorithm for apparent diseases of reinforced concrete structures bridge based on improved YOLOv8s[J]. Science Technology and Engineering, 2026, 26(4): 1676-1687.
SongJuan, HeLongxi, LongHuiping. Deep learning-based algorithm for multi defectdetection in tunnel lining[J]. Journal of Zhejiang University (Engineering Science Edition), 2024, 58(6): 1161-1173.
[14]
WuX, GuoN. MGSLU-Net: a lightweight network for efficient detection of water leakage in subway tunnel linings[J]. The Visual Computer, 2025, 41(10): 1-14.
TianSimeng, WangYaodong, LiuShengzhi, et al. Research on crack and leakage image acquisition and intelligent recognition algorithm of subway shield tunnel[J]. Journal of Central South University (Science and Technology Edition), 2025, 56(1): 370-383.
LiYong, LinFang, ChenYuang, et al. A comprehensive review of deep learning-based visual multi-object tracking[J]. Science Technology and Engineering, 2025, 25(22): 9211-9223.
ZhaoZiqi, LiWeidong, LiXiaojuan. Small target detection algorithm in aerialimages based on improved RT-DETR[J]. Science Technology and Engineering, 2025, 25(13): 5527-5534.
JiangShixin, TanChuncheng, YangJianxi, et al. Lightweight detection method of surface damage on concrete bridges based on improved segformer[J]. China Journal of Highway and Transport, 2024, 37(2): 77-87.
WangYanwei, TaoWenbin, ChenKaiyun, et al. Coal gangue recognition technology based on SegFormer-CG[J]. Journal of China Coal Society, 2026: 51(4): 2771-2782.
[29]
DengX Y, ZhengY P. GLF-Segformer: an improved Segformer model integrating local and global information for skin cancer ima-ge segmentation[J]. Optoelectronics Letters, 2026, 22(4): 236-242.
WenTiantian, PuYunwei, ZhaoWenxiang. Method for sea-land segmentation combining contextual semantic information and edge features[J]. Remote Sensing for Natural Resources, 2025, 37(5): 62-72.
[38]
XieE, WangW, YuZ, et al. SegFormer: simple and efficient design for semantic segmentation with transformers[J]. Advances in Neural Information Processing Systems, 2021, 34: 12077-12090.
XiongZhifei, ShenJianghai. A multi-focus image fusion algorithm based on multidimensional adaptive filtering and poisson fusion[J]. Science Technology and Engineering, 2023, 23(24): 10427-10436.
HouMingming, NiuXiaochuan, WangKexun. Detection algorithm for protective equipment of railway workers based on improved YOLOv5[J]. Science Technology and Engineering, 2025, 25(23): 9859-9865.
[43]
YuanX, ZhengZ, LiY, et al. Strip R-CNN: large strip convolution for remote sensing object detection[J]. arXiv Preprint arXiv: 2501.03775.
[44]
LiM, ChenM, LuW, et al. Automatic extraction and quantitative analysis of characteristics from complex fractures on rock surfaces via deep learning[J]. International Journal of Rock Mecha-nics and Mining Sciences, 2025, 187: 106038.
LiHaoyu, LiaoWeizhan. Building component point cloud extraction method based on improved RandLA-Net[J]. Science Technology and Engineering, 2025, 25(6): 2461-2468.
LuoYaowei, LiuAirong, ChenBingcong, et al. Bridge underwater structural crack detection based on image enhancement and improved U-net fusion[J]. Engineering Mechanics, 2025, 42(S1): 276-282, 306.
[49]
XueY, CaiX, ShadabfarM, et al. Deep learning-based automa-tic recognition of water leakage area in shield tunnel lining[J]. Tunnelling and Underground Space Technology, 2020, 104: 103524.