隧道工程岩体节理地质注意力YOLO目标检测算法

李嘉豪; 张中霖; 申华远; 付子鸣; 吴兴扬; 白若彤; 楚梓苹

doi:10.19509/j.cnki.dzkq.tb20250463

隧道工程岩体节理地质注意力YOLO目标检测算法

doi: 10.19509/j.cnki.dzkq.tb20250463

a.
西安建筑科技大学：建筑设备科学与工程学院
b.
西安建筑科技大学：安德学院，西安 710055

详细信息

作者简介:
李嘉豪：E-mail：19591503629@163.com

通讯作者:
E-mail：zhang_zhonglin215@xauat.edu.cn

中图分类号: TP391.41；TU456.3；TU457
计量
- 文章访问数: 190
- PDF下载量: 7
- 被引次数: 0
出版历程
- 收稿日期: 2025-10-26
- 录用日期: 2026-03-27
- 修回日期: 2026-03-23
- 网络出版日期: 2026-03-27

Research on geology-specific attention-based YOLO object detection algorithm for rock joints in tunnel engineering

a.
School of Building Services Science and Engineering, Xi'an University of Architecture and Technology
b.
An De College, Xi'an University of Architecture and Technology, Xi'an 710055, China

More Information

Author Bio:
E-mail：19591503629@163.com

Corresponding author: E-mail：zhang_zhonglin215@xauat.edu.cn

摘要

摘要:
隧道工程岩体节理检测是确保隧道结构稳定的关键一环。目前主要的隧道检测方法受人为主观影响存在漏检、误检率高等问题，现有深度学习检测模型对细微节理的全局定位与特征捕捉能力不足。本研究提出一种地质注意力 YOLO 目标检测算法 GeoLA-YOLO：在 YOLOv8n 骨干网络嵌入 CBAM 注意力机制构建 C2f_CBAM 增强模块，提升模型对细微特征的提取能力；对检测头采用分类−回归解耦，搭配 Dropout、组归一化等轻量化正则化策略，增强对细微节理的精准定位与判别能力。在自建数据集中的实验结果表明，改进后的算法在参数量方面与基线 YOLOv8n 算法保持相近的同时，在 mAP@0.5、mAP@0.5:0.95、Recall、F1分数指标上分别提高了 4.3%，9.6%，5.0%，5.5 %，验证了模型的有效性。在公开数据集 CUBIT-Det 中实验结果表明，改进后的模型与基线 YOLOv8n 算法相比在 mAP@0.5、mAP@0.5：0.95 指标上分别提升 6.2%，5.9%，验证了 GeoLA-YOLO 的鲁棒性。GeoLA-YOLO 算法在保持轻量化的同时有效提升了隧道岩体节理检测的精度与鲁棒性，为隧道工程岩体节理的智能精准检测提供了实用技术支撑，可以满足隧道施工安全智能化监测需求。
- 隧道工程岩体 /
- 隧道掌子面节理 /
- GeoLA-YOLO /
- YOLOv8n /
- 注意力机制 /
- 细微特征
Abstract:
Objective
Rock joint detection in tunnel engineering is a critical component for ensuring tunnel structural stability. Current tunnel detection methods are susceptible to human subjectivity, resulting in high missed-detection and false-detection rates. Existing deep learning-based detection models exhibit insufficient performance in global localization and feature capturing for subtle rock joints.
Methods
To address these challenges, this study proposed GeoLA-YOLO, a geology-specific attention-based YOLO object detection algorithm. The framework incorporated the CBAM attention mechanism into the backbone network of YOLOv8n to construct the C2f_CBAM enhanced module, thereby improving the model's ability to extract subtle rock-joint features. Furthermore, classification–regression decoupling was adopted for the detection head, combined with lightweight regularization strategies including Dropout and group normalization, to improve the precise localization and discrimination of subtle joints.
Results
Experimental results on the self-built VOC (visual object classes) dataset showed that the optimized algorithm maintained lightweight performance while improving mAP@0.5, mAP@0.5:0.95, Recall, and F1 metrics by 4.3, 9.6, 5.0, and 5.5 percentage points, respectively, compared to the original algorithm, demonstrating the model's effectiveness. Experimental results on the public dataset CUBIT-Det showed that the improved model outperformed the baseline YOLOv8n algorithm, with mAP@0.5 and mAP@0.5:0.95 increased by 6.2 and 5.9 percentage points, respectively. This confirmed the robustness of GeoLA-YOLO.
Conclusion
The GeoLA-YOLO algorithm effectively enhances the accuracy and robustness of tunnel rock joint detection under the premise of maintaining lightweight properties and adapting to resource-limited on-site detection devices for tunnel engineering. It provides practical technical support for intelligent and precise detection of rock joints in tunnel engineering and meets the requirements for intelligent safety monitoring during tunnel construction.
- tunnel rock mass /
- tunnel face joint /
- GeoLA-OLO /
- YOLOv8n /
- attention mechanism /
- subtle features

HTML全文

图 1 C2f_CBAM注意力增强模块结构

Conv为卷积层；Split为特征分割；Bottleneck为瓶颈单元；Concat为特征拼接；CBAM为卷积块注意力模块，由通道注意力模块与空间注意力模块串联构成；⊕表示残差相加；⊗表示特征加权相乘；Backnone为主干网络；H，W，C分别为特征图的高度、宽度、通道数；k，s，p分别为卷积核大小、步长、填充；C2f为YOLOv8的特征提取模块；C2f_CBAM为嵌为CBAM注意力的C2f模块；SPPE为空间金字塔池化快速版；下同

Figure 1. Structure of C2f_CBAM Attention-Enhanced Module

下载: 全尺寸图片幻灯片

图 2 GeoLA-YOLO网络结构框架

Neck为特征融合颈部网络；Head为检测头，最终检测输出通过Detect模块实现，分别输出边界框回归与类别分类结果；Upsample为最近邻上采样；Dropout为随机失活层，用于抑制过拟合；Improved Module为改进模块；Group Norm为组归一化；Bbox. Loss为边界框损失；Cls. Loss为分类损失；Conv2d为二维卷积层；下同

Figure 2. GeoLA-YOLO network structure framework

下载: 全尺寸图片幻灯片

图 3 注意力机制模块

CBAM由通道注意力模块与空间注意力模块串联组成；通道注意力利用全局平均池化与全局最大池化聚合特征，经多层感知机（MLP）生成通道权重；空间注意力对池化拼接特征进行卷积得到空间权重；MaxPool，AvgPool为最大池化和平均池化；GAP为全局平均池化；GMP为全局最大池化

Figure 3. Attention mechanism module

下载: 全尺寸图片幻灯片

图 4 检测头模块Detect结构框架

检测头在输入前端引入Dropout抑制过拟合，分为边界框回归分支与分类分支；bias=False表示卷积无偏置

Figure 4. Detect structure framework

下载: 全尺寸图片幻灯片

图 5 掌子面围岩图像数据空间增强

Figure 5. Spatial enhancement of image data of surrounding rock at tunnel face

下载: 全尺寸图片幻灯片

图 6 数据集分布示意

Figure 6. Schematic diagram of dataset distribution

下载: 全尺寸图片幻灯片

图 7 应用Labelme软件进行标注

Figure 7. Annotation using Labelme software

下载: 全尺寸图片幻灯片

图 8 检测效果可视化

Figure 8. Visualization of detection results

下载: 全尺寸图片幻灯片

图 9 GeoLA-YOLO与YOLOv8训练损失、精度与召回率对比

Figure 9. Comparison of training loss, precision and recall between GeoLA-YOLO and YOLOv8

下载: 全尺寸图片幻灯片

表 1 实验环境配置

Table 1. Experimental environment settings

实验环境	版本型号
CPU	Intel(R) Core(TM) i7-14700HX
GPU	NVIDIA RTX 4070显卡
Operating System	Windows11（64位）
CUDA	11.6

下载: 导出CSV

表 2 实验超参数配置

Table 2. Experimental hyperparameter settings

超参数类别	参数配置
迭代次数（Epochs）	350
训练图片尺寸（Image Size）	640×640
批处理大小（Batch Size）	16
优化器（Optimizer）	Adam
初始学习率（Learning Rate）	0.0003
权重衰减系数（Weight decay）	0.001

下载: 导出CSV

表 3 GeoLA-YOLO消融实验结果

Table 3. GeoLA-YOLO ablation experiment results

实验组	FPS/(帧·s⁻¹)	参数量/10⁶	训练耗时/h	mAP@0.5/%	mAP@0.5-0.95/%	Recall/%	F1-Score/%
原始组	140	3.05	5.40	91.8	58.9	87.0	88.5
无CBAM注意力组	135.1	3.20	5.05	92	61.3	89.1	87.5
通道精简组	74.9	11.21	5.39	96.7	65.5	95.6	92.6
深度优化组	74.3	11.22	4.90	95.7	64.9	91.5	91.9
无Dropout组	117.6	3.20	4.08	91.1	59.6	88.7	86.8
完整改进组	137.0	3.20	4.06	96.1	68.5	92.0	94.0
注：FPS. 帧率；mAP@0.5为交并比IoU阈值未0.5时的目标检测精度；mAP@0.5:0.95为不同IoU阈值（0.5～0.95，间隔0.05共10个阈值）下mAP的平均值；Recall为召回率；F1-Score为F1分数；下同

下载: 导出CSV

表 4 对比实验结果

Table 4. Comparative experiment results

模型	参数量	mAP@0.5/%	mAP@0.5:0.95/%
YOLOv5-n^[39]	1.76	73.4	39.9
YOLOv7-t^[40]	6.01	71.1	39.7
YOLOX-n^[41]	2.24	73.0	39.5
YOLOv8n	3.04	69.4	42.9
PP-YOLOE-s	8.02	64.6	38.9
PP-YOLOE+-s	8.02	70.6	44.0
Faster R-CNN (MBNv2)	19.36	45.6	19.4
Faster R-CNN^[42]	42.62	71.5	43.3
Faster R-CNN (Res50)	42.62	71.5	43.3
GeoLA-YOLO	3.20	75.6%	48.8%

下载: 导出CSV

参考文献(42)

[1]	巩江峰, 王伟, 黎旭, 等. 截至 2024 年底中国铁路隧道情况统计及 2024 年新开通项目隧道情况介绍[J]. 隧道建设(中英文), 2025, 45(3): 636-653. GONG J F, WANG W, LI X, et al. Statistics of China's railway tunnels by the end of 2024 and overview of tunnels of new projects in 2024[J]. Tunnel Construction, 2025, 45(3): 636-653. (in Chinese with English abstract)
[2]	宋战平, 杨子凡, 张玉伟, 等. 深度学习在隧道与地下工程中的应用现状及展望[J]. 隧道建设(中英文), 2025, 45(2): 221-255. SONG Z P, YANG Z F, ZHANG Y W, et al. Current status and prospects of deep learning applications in tunnel and underground engineering[J]. Tunnel Construction, 2025, 45(2): 221-255. (in Chinese with English abstract)
[3]	陈志新, 肖勇卓, 蔡永昌, 等. 基于图机器视觉的山岭隧道掌子面节理裂隙检测方法研究进展综述[J]. 现代隧道技术, 2025, 62(5): 1-12. doi: 10.13807/j.cnki.mtt.2025.05.001 CHEN Z X, XIAO Y Z, CAI Y C, et al. Review of research progress on machine vision-based joint and fracture detection for mountain tunnel faces[J]. Modern Tunnelling Technology, 2025, 62(5): 1-12. (in Chinese with English abstract) doi: 10.13807/j.cnki.mtt.2025.05.001
[4]	魏培勇, 金晓武, 焦利, 等. 铁路无人机巡检技术研究与应用[J]. 铁道建筑, 2023, 63(4): 108-113. doi: 10.3969/j.issn.1003-1995.2023.04.25 WEI P Y, JIN X W, JIAO L, et al. Research and application of railway UAV patrol inspection technology[J]. Railway Engineering, 2023, 63(4): 108-113. (in Chinese with English abstract) doi: 10.3969/j.issn.1003-1995.2023.04.25
[5]	叶英, 胡刚, 范留明. 传统数字图像处理在隧道掌子面节理提取中的应用局限 [J]. 岩土工程技术, 2025, 39(3): 189-196. YE Y, HU G, FAN L M. Application limitations of traditional digital image processing in joint extraction of tunnel face[J]. Geotechnical Engineering Technology, 2025, 39(3): 189-196 (in Chinese with English abstract)
[6]	郭延辉, 李顺银, 严航, 等. 深埋隧道穿越富水断层破碎带突水突泥灾变演化规律研究[J]. 地质科技通报, 2026, 45(1): 160-170. doi: 10.19509/j.cnki.dzkq.tb20240160 GUO Y H, LI S Y, YAN H, et al. Evolution law of water and mud inrush disaster in deep-buried tunnel crossing water-rich fault fracture zone[J]. Bulletin of Geological Science and Technology, 2026, 45(1): 160-170. (in Chinese with English abstract) doi: 10.19509/j.cnki.dzkq.tb20240160
[7]	ZHOU Z, LI S S, YAN L B, et al. Intelligent recognition of tunnel lining defects based on deep learning: Methods, challenges and prospects[J]. Engineering Failure Analysis, 2025, 170: 109332. doi: 10.1016/j.engfailanal.2025.109332
[8]	QIN H, ZHANG D H, TANG Y, et al. Automatic recognition of tunnel lining elements from GPR images using deep convolutional networks with data augmentation[J]. Automation in Construction, 2021, 130: 103830. doi: 10.1016/j.autcon.2021.103830
[9]	WANG J, ZHANG J Q, COHN A G, et al. Arbitrarily-oriented tunnel lining defects detection from ground penetrating radar images using deep convolutional neural networks[J]. Automation in Construction, 2022, 133: 104044. doi: 10.1016/j.autcon.2021.104044
[10]	岩峰, 王浩浩, 胡永立, 等. 布里渊光时域传感技术在隧道监测中的关键问题及应用现状[J]. 地质科技通报, 2026, 45(3): 103-114. doi: 10.19509/j.cnki.dzkq.tb20250269 YAN F, WANG H H, HU Y L, et al. Brillouin optical time-domain sensing technology and its application in tunnel monitoring[J]. Bulletin of Geological Science and Technology, 2026, 45(3): 103-114. (in Chinese with English abstract) doi: 10.19509/j.cnki.dzkq.tb20250269
[11]	ZHAO L, LI J H, SONG Z P, et al. A collaborative YOLOv8n model for joint detection in complex tunnel environments using attention and super-resolution[J]. Engineering Research Express, 2026, 8(6): 065105. doi: 10.1088/2631-8695/ae4f5c
[12]	郭延辉, 罗毅, 刘敬, 等. 乌蒙山区特长极深隧道隧址区地应力场分布特征及反演分析[J]. 地质科技通报, 2026, 45(5): 159-170. GUO Y H, LUO Y, LIU J, et al. Distribution characteristics and inversion analysis of in-situ stress field at tunnel site of an extra-long and ultra-deep tunnel in Wumeng Mountain area[J]. Bulletin of Geological Science and Technology, 2026, 45(5): 159-170. (in Chinese with English abstract)
[13]	谌婷婷, 魏怡. 基于改进YOLOv4的混凝土裂缝检测方法[J]. 激光杂志, 2024, 45(1): 80-85. doi: 10.14016/j.cnki.jgzz.2024.1.080 CHEN T T, WEI Y. Improved YOLOv4-based concrete crack detection method[J]. Laser Journal, 2024, 45(1): 80-85. (in Chinese with English abstract) doi: 10.14016/j.cnki.jgzz.2024.1.080
[14]	彭磊, 周春, 胡锋, 等. 基于深度学习的隧道掌子面节理智能检测与分割[J]. 人民长江, 2023, 54(10): 243-250. doi: 10.16232/j.cnki.1001-4179.2023.10.035 PENG L, ZHOU C, HU F, et al. Intelligent detection and segmentation of tunnel face joints based on deep learning[J]. Yangtze River, 2023, 54(10): 243-250. (in Chinese with English abstract) doi: 10.16232/j.cnki.1001-4179.2023.10.035
[15]	罗虎, MARK M, 张睿, 等. 基于计算机视觉技术和深度学习的隧道掌子面岩体裂隙自动识别方法研究[J]. 现代隧道技术, 2023, 60(1): 56-65. doi: 10.13807/j.cnki.mtt.2023.01.006 LUO H, MARK M, ZHANG R, et al. Research on the automatic identification method for rock mass fracture in tunnel face based on computer vision technology and deep learning[J]. Modern Tunnelling Technology, 2023, 60(1): 56-65. (in Chinese with English abstract) doi: 10.13807/j.cnki.mtt.2023.01.006
[16]	FLOOD I, KARTAM N. Neural networks in civil engineering. I: Principles and understanding[J]. Journal of Computing in Civil Engineering, 1994, 8(2): 131-148. doi: 10.1061/(ASCE)0887-3801(1994)8:2(131)
[17]	WANG S H, NI P P, GUO M D. Spatial characterization of joint planes and stability analysis of tunnel blocks[J]. Tunnelling and Underground Space Technology, 2013, 38: 357-367. doi: 10.1016/j.tust.2013.07.017
[18]	HE P, LI S C, LI L P, et al. Discontinuous deformation analysis of super section tunnel surrounding rock stability based on joint distribution simulation[J]. Computers and Geotechnics, 2017, 91: 218-229. doi: 10.1016/j.compgeo.2017.07.022
[19]	FAN H Y, LI L P, ZONG P J, et al. Advanced stability analysis method for the tunnel face in jointed rock mass based on DFN-DEM[J]. Underground Space, 2023, 13: 136-149. doi: 10.1016/j.undsp.2023.03.009
[20]	CHEN J Y, ZHOU M L, HUANG H W, et al. Automated extraction and evaluation of fracture trace maps from rock tunnel face images via deep learning[J]. International Journal of Rock Mechanics and Mining Sciences, 2021, 142: 104745. doi: 10.1016/j.ijrmms.2021.104745
[21]	CHEN J Y, YANG T J, ZHANG D M, et al. Deep learning based classification of rock structure of tunnel face[J]. Geoscience Frontiers, 2021, 12(1): 395-404. doi: 10.1016/j.gsf.2020.04.003
[22]	CHEN J Y, HUANG H W, ZHOU M L, et al. Towards semi-automatic discontinuity characterization in rock tunnel faces using 3D point clouds[J]. Engineering Geology, 2021, 291: 106232. doi: 10.1016/j.enggeo.2021.106232
[23]	CHEN J Y, HUANG H W, COHN A G, et al. Machine learning-based classification of rock discontinuity trace: SMOTE oversampling integrated with GBT ensemble learning[J]. International Journal of Mining Science and Technology, 2022, 32(2): 309-322. doi: 10.1016/j.ijmst.2021.08.004
[24]	CHEN J Y, CHEN Y F, COHN A G, et al. A novel image-based approach for interactive characterization of rock fracture spacing in a tunnel face[J]. Journal of Rock Mechanics and Geotechnical Engineering, 2022, 14(4): 1077-1088. doi: 10.1016/j.jrmge.2021.10.012
[25]	LEI M F, ZHANG Y B, DENG E, et al. Intelligent recognition of joints and fissures in tunnel faces using an improved mask region-based convolutional neural network algorithm[J]. Computer-Aided Civil and Infrastructure Engineering, 2024, 39(8): 1123-1142. doi: 10.1111/mice.13097
[26]	ZHAO L, HAO S Y, SONG Z P. Context-aware semantic segmentation network for tunnel face feature identification[J]. Automation in Construction, 2024, 165: 105560. doi: 10.1016/j.autcon.2024.105560
[27]	KIM Y, YUN T S. Enhanced rock mass rating prediction from tunnel face imagery: A decision-supportive ensemble deep learning approach[J]. Engineering Geology, 2024, 339: 107625. doi: 10.1016/j.enggeo.2024.107625
[28]	PHAM C, KIM B C, SHIN H S. Deep learning-based identification of rock discontinuities on 3D model of tunnel face[J]. Tunnelling and Underground Space Technology, 2025, 158: 106403. doi: 10.1016/j.tust.2025.106403
[29]	JI Y P, SONG S Y, ZHANG W, et al. Automatic identification of rock fractures based on deep learning[J]. Engineering Geology, 2025, 345: 107874. doi: 10.1016/j.enggeo.2024.107874
[30]	LI M Z, CHEN M, LU W B, et al. Automatic extraction and quantitative analysis of characteristics from complex fractures on rock surfaces via deep learning[J]. International Journal of Rock Mechanics and Mining Sciences, 2025, 187: 106038. doi: 10.1016/j.ijrmms.2025.106038
[31]	ZHOU Z, LI H C, ZHOU S R, et al. A deep learning-based algorithm for fast identification of multiple defects in tunnels[J]. Engineering Applications of Artificial Intelligence, 2025, 145: 110035. doi: 10.1016/j.engappai.2025.110035
[32]	ZHAO S, ZHANG D M, HUANG H W. Deep learning–based image instance segmentation for moisture marks of shield tunnel lining[J]. Tunnelling and Underground Space Technology, 2020, 95: 103156. doi: 10.1016/j.tust.2019.103156
[33]	ZHAO S, ZHANG D M, XUE Y D, et al. A deep learning-based approach for refined crack evaluation from shield tunnel lining images[J]. Automation in Construction, 2021, 132: 103934. doi: 10.1016/j.autcon.2021.103934
[34]	ZHAO S, ZHANG G K, ZHANG D M, et al. A hybrid attention deep learning network for refined segmentation of cracks from shield tunnel lining images[J]. Journal of Rock Mechanics and Geotechnical Engineering, 2023, 15(12): 3105-3117. doi: 10.1016/j.jrmge.2023.02.025
[35]	ZHANG Z L, CHEN D F, YANG J Y, et al. Accurate detection of photovoltaic panel defects via visible-infrared image fusion and a lightweight enhanced YOLOv11n model[J]. Engineering Research Express, 2026, 8(1): 015234. doi: 10.1088/2631-8695/ae3457
[36]	WOO S, PARK J, LEE J Y, et al. CBAM: Convolutional block attention module[C]//Computer Vision – ECCV 2018. Cham: Springer, 2018: 3-19.
[37]	谢雄耀, 王皓正, 周彪, 等. 复杂环境公路隧道裂缝快速识别与分割算法研究[J]. 地下空间与工程学报, 2022, 18(3): 1025-1033. doi: 10.20174/j.juse.2022.03.035 XIE X Y, WANG H Z, ZHOU B, et al. Research on fast identification and segmentation algorithms for cracks of highway tunnels in complex environment[J]. Chinese Journal of Underground Space and Engineering, 2022, 18(3): 1025-1033. (in Chinese with English abstract) doi: 10.20174/j.juse.2022.03.035
[38]	ZHAO B Y, ZHOU X K, YANG G D, et al. High-resolution infrastructure defect detection dataset sourced by unmanned systems and validated with deep learning[J]. Automation in Construction, 2024, 163: 105405. doi: 10.1016/j.autcon.2024.105405
[39]	JOCHER G. YOLOv5 by Ultralytics[EB/OL]. https://github.com/ultralytics/yolov5, 2020.
[40]	WANG C Y, BOCHKOVSKIY A, LIAO H M. YOLOv7: Trainable bag-of-freebies sets new state-of-the-art for real-time object detectors[C]//2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Vancouver, BC, Canada. IEEE, 2023: 7464-7475.
[41]	GE Z, LIU S T, WANG F, et al. YOLOX: Exceeding YOLO series in 2021[J]. https://doi.org/10.48550/arXiv.2107.08430, 2021.
[42]	REN S Q, HE K M, GIRSHICK R, et al. Faster R-CNN: Towards real-time object detection with region proposal networks[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017, 39(6): 1137-1149. doi: 10.1109/TPAMI.2016.2577031