• ISSN 0258-2724
  • CN 51-1277/U
  • EI Compendex
  • Scopus 收录
  • 全国中文核心期刊
  • 中国科技论文统计源期刊
  • 中国科学引文数据库来源期刊

基于轻量级模型的隧道岩性快速识别方法

夏毅敏 李清友 邓朝辉 龙斌 姚捷

夏毅敏, 李清友, 邓朝辉, 龙斌, 姚捷. 基于轻量级模型的隧道岩性快速识别方法[J]. 西南交通大学学报, 2021, 56(2): 420-427. doi: 10.3969/j.issn.0258-2724.20191057
引用本文: 夏毅敏, 李清友, 邓朝辉, 龙斌, 姚捷. 基于轻量级模型的隧道岩性快速识别方法[J]. 西南交通大学学报, 2021, 56(2): 420-427. doi: 10.3969/j.issn.0258-2724.20191057
XIA Yimin, LI Qingyou, DENG Chaohui, LONG Bin, YAO Jie. Rapid Identification Method for Lithology of Tunnel Based on Lightweight Model[J]. Journal of Southwest Jiaotong University, 2021, 56(2): 420-427. doi: 10.3969/j.issn.0258-2724.20191057
Citation: XIA Yimin, LI Qingyou, DENG Chaohui, LONG Bin, YAO Jie. Rapid Identification Method for Lithology of Tunnel Based on Lightweight Model[J]. Journal of Southwest Jiaotong University, 2021, 56(2): 420-427. doi: 10.3969/j.issn.0258-2724.20191057

基于轻量级模型的隧道岩性快速识别方法

doi: 10.3969/j.issn.0258-2724.20191057
基金项目: 国家重点研发计划资助项目(2017YFB1302600);湖南省科技重大专项(2019GK1010)
详细信息
    通讯作者:

    夏毅敏(1987—),男,教授,研究方向为地下工程与大型掘进装备,E-mail:xiamj@csu.edu.cn

  • 中图分类号: V221.3

Rapid Identification Method for Lithology of Tunnel Based on Lightweight Model

  • 摘要: 为了解决隧道岩性现有识别方法中识别时间长、安全性低、主观性大等问题,结合不同岩性表面具有不同的成分特征,提出了一种基于轻量级模型与岩石图像的隧道岩性快速识别方法. 首先,通过相机采集隧道常见的片麻岩、花岗岩、石灰岩、大理岩、凝灰岩、砂岩等6类主要岩石,建立了岩石图像数据集并划分训练集、验证集与测试集;然后,基于轻量级模型MobileNet V2在ImageNet数据集上进行预训练,改进模型分类器结构以适应岩石数据集,并采用模型迁移学习训练方法对1170张训练集图像进行训练,获取了岩石岩性识别模型;最后,选取共计300张测试集图像在离线条件下进行了模型测试,并与VGG16模型与SVM (support vector machine)模型进行了对比. 实验结果表明:模型在测试数据集上的各项总体评估指标均在85%以上,其中凝灰岩各项评价指标达到94%以上,模型大小仅28.3 MB,平均识别时间为2880 ms,表明该识别模型体积小,识别准确率高,识别时间快,在精确率与识别速度上均优于传统方法.

     

  • 图 1  标准卷积结构

    Figure 1.  Standard convolution structure

    图 2  深度可分离式卷积结构

    Figure 2.  Structure of depthwise separable convolution

    图 3  传统学习与迁移学习

    Figure 3.  Traditional learning and transfer learning

    图 4  模型训练与识别流程

    Figure 4.  Model training and identification process

    图 5  岩石样本图像示例

    Figure 5.  Rock sample image examples

    图 6  MobileNet V2深度可分离式卷积结构

    Figure 6.  MobileNet V2 depthwise separable convolution structure

    图 7  训练中训练精度、验证精度及损失变化

    Figure 7.  Training accuracy,verification accuracy and loss changes during training

    表  1  MobileNet V2基础结构

    Table  1.   MobileNet V2 basic structure

    输入操作tcns
    2242 × 3conv2d3212
    1122 × 32bottleneck11611
    1122 × 16bottleneck62422
    562 × 24bottleneck63232
    282 × 32bottleneck66442
    142 × 64bottleneck69631
    142 × 96bottleneck616032
    72 × 160bottleneck632011
    72 × 320conv2d 1 × 1128011
    72 × 1280avgpool 7 × 71
    1 × 1 × 1280conv2d1000
    注:c 为输出通道数;s 为模块第 1 次重复时的操作步长(之后重复操作步长均为 1).
    下载: 导出CSV

    表  2  改进的模型结构

    Table  2.   Improved model structure

    输入操作输出
    224 × 224 × 3conv2d112 × 112 × 32
    112 × 112 × 32bottleneck1112 × 112 × 16
    $ \vdots $$\vdots $$\vdots $
    7 × 7 × 160bottleneck167 × 7 × 160
    7 × 7 × 160bottleneck177 × 7 × 320
    7 × 7 × 320conv2d 1 × 17 × 7 × 1280
    7 × 7 × 1280avgpool 7 × 71 × 1 × 1280
    1 × 1 × 1280Dense1 (激活函数 Relu)1 × 1 × 1024
    1 × 1 × 1024Dropout1 × 1 × 1024
    1 × 1 × 1024Dense2 (激活函数 Relu)1 × 1 × 256
    1 × 1 × 256Dropout1 × 1 × 256
    1 × 1 × 256Dense3 (激活函数 Relu)1 × 1 × 6
    1 × 1 × 6Softmax 分类函数分类结果
    下载: 导出CSV

    表  3  基于fine-tune的迁移训练

    Table  3.   Transfer training based on fine-tune

    输入操作输出
    224 × 224 × 3conv2d112 × 112 × 32
    112 × 112 × 32bottleneck1112 × 112 × 16
    $ \vdots $$\vdots $$\vdots $
    7 × 7 × 160bottleneck167 × 7 × 160
    7 × 7 × 160bottleneck177 × 7 × 320
    7 × 7 × 320conv2d 1 × 17 × 7 × 1280
    7 × 7 × 1280avgpool 7 × 71 × 1 × 1280
    1 × 1 × 1280Dense (激活函数 Relu)1 × 1 × 1024
    1 × 1 × 1024Dropout1 × 1 × 1024
    1 × 1 × 1024Dense (激活函数 Relu)1 × 1 × 256
    1 × 1 × 256Dropout1 × 1 × 256
    1 × 1 × 256Dense (激活函数 Relu)1 × 1 × 6
    1 × 1 × 6Softmax 分类函数分类结果
    下载: 导出CSV

    表  4  训练集各项评价指标

    Table  4.   Index values for training set evaluation

    岩石精确率召回率综合评价指标
    片麻岩1.0000.9330.965
    花岗岩1.0000.9670.983
    大理岩0.9381.0000.968
    石灰岩0.9350.9670.951
    凝灰岩0.9681.0000.984
    砂岩1.0000.9670.983
    下载: 导出CSV

    表  5  测试集图片识别结果

    Table  5.   Recognition results for testing set images

    测试图片
    片麻岩花岗岩大理岩石灰岩凝灰岩砂岩
    片麻岩362561
    花岗岩4514
    大理岩1463
    石灰岩4451
    凝灰岩248
    砂岩4314236
    合计405157645137
    下载: 导出CSV

    表  6  测试集各项评价指标

    Table  6.   Index values for testing set evaluation

    岩石PRF1
    片麻岩0.9000.7200.800
    花岗岩0.8820.9000.891
    大理岩0.8070.9200.860
    石灰岩0.7030.9000.789
    凝灰岩0.9410.9600.950
    砂岩0.9730.7200.828
    下载: 导出CSV

    表  7  实验结果对比

    Table  7.   Comparison of experimental results

    分类指标平均识别精确率单张识别时间/ms
    SVM0.6654650
    VGG160.8143680
    MobileNet V20.8682880
    下载: 导出CSV
  • 管会生,吴和北,陶伟,等. 深埋斜井隧道双模式盾构刀盘驱动扭矩分析[J]. 现代隧道技术,2015,52(6): 163-169.

    GUAN Huisheng, WU Hebei, TAO Wei, et al. Analysis of the cutterhead driving torque of a dual-mode shield used for a deeply buried inclined shaft[J]. Modern Tunnelling Technology, 2015, 52(6): 163-169.
    刘书斌, 方勇, 周超月, 等. 承压溶腔对隧道开挖稳定性影响的模型试验[J]. 岩石力学与工程学报, 2017, 36(增刊2): 3781-3791.

    LIU Shubin, FANG Yong, ZHOU Chaoyue, et al. Model test of tunnel excavation stability influenced by concealed cave with internal water pressure[J]. Chinese Journal of Rock Mechanics and Engineering, 2017, 36(S2): 3781-3791.
    严鹏,张晨,高启栋,等. 不同损伤程度下岩石力学参数变化的声波测试[J]. 岩土力学,2015,36(12): 3425-3432.

    YAN Peng, ZHANG Chen, GAO Qidong, et al. Acoustic wave test on mechanical properties variation of rocks under different damage degrees[J]. Rock and Soil Mechanics, 2015, 36(12): 3425-3432.
    苏凯,张妍珺,伍鹤皋,等. 隧洞开挖过程中围岩安全系数演化特征与支护时机选择方法研究[J]. 岩石力学与工程学报,2019,38(增刊1): 2964-2975.

    SU Kai, ZHANG Yanjun, WU Hegao, et al. Evolution of surrounding rock safety factor and support installation time during tunnel excavation[J]. Chinese Journal of Rock Mechanics and Engineering, 2019, 38(S1): 2964-2975.
    WEI X S, QIN X H, RONG C L, et al. Image classification recognition for rock micro-thin section based on probabilistic neural networks[J]. Applied Mechanics and Materials, 2014, 602-605: 2147-2152. doi: 10.4028/www.scientific.net/AMM.602-605.2147
    罗佳,刘大刚. 基于自适应阈值和连通域的隧道裂缝提取[J]. 西南交通大学学报,2018,53(6): 1137-1141. doi: 10.3969/j.issn.0258-2724.2018.06.007

    LUO Jia, LIU Dagang. Tunnel crack extraction based on adaptive threshold and connected domain[J]. Journal of Southwest Jiaotong University, 2018, 53(6): 1137-1141. doi: 10.3969/j.issn.0258-2724.2018.06.007
    刘烨,程国建,马微,等. 基于铸体薄片图像颜色空间与形态学梯度的岩石分类[J]. 中南大学学报(自然科学版),2016,47(7): 2375-2382.

    LIU Ye, CHENG Guojian, MA Wei, et al. Rock classification based on features form color space and morphological gradient of rock thin section image[J]. Journal of Central South University (Science and Technology), 2016, 47(7): 2375-2382.
    程国建,郭文惠,范鹏召. 基于卷积神经网络的岩石图像分类[J]. 西安石油大学学报(自然科学版),2017,32(4): 116-122. doi: 10.3969/j.issn.1673-064X.2017.04.020

    CHENG Guojian, GUO Wenhui, FAN Pengshao. Study on rock image classification based on convolution neural network[J]. Journal of Xi’an Shiyou University (Natural Science Edition), 2017, 32(4): 116-122. doi: 10.3969/j.issn.1673-064X.2017.04.020
    LECUN Y, BENGIO Y, HINTON G. Deep learning[J]. Nature, 2015, 521: 436-444. doi: 10.1038/nature14539
    BIANCO S, BUZZELLI M, MAZZINI D, et al. Deep learning for logo recognition[J]. Neurocomputing, 2017, 245: 23-30. doi: 10.1016/j.neucom.2017.03.051
    侯进,吕志良,徐茂,等. 基于深度学习的复合神经网络机场信号检测框架[J]. 西南交通大学学报,2019,54(4): 863-869.

    HOU Jin, LYU Zhiliang, XU Mao, et al. Combined neural networks based on deep learning for signal detection in aeronautical communications[J]. Journal of Southwest Jiaotong University, 2019, 54(4): 863-869.
    白林,姚钰,李双涛,等. 基于深度学习特征提取的岩石图像矿物成分分析[J]. 中国矿业,2018,27(7): 178-182.

    BAI Lin, YAO Yu, LI Shuangtao, et al. Mineral composition analysis of rock image based on deep learning feature extraction[J]. China Mining Magazine, 2018, 27(7): 178-182.
    徐述腾,周永章. 基于深度学习的镜下矿石矿物的智能识别实验研究[J]. 岩石学报,2018,34(11): 3244-3252.

    XU Shuteng, ZHOU Yongzhang. Artificial intelligence identification of ore minerals under microscope based on deep learning algorithm[J]. Acta Petrologica Sinica, 2018, 34(11): 3244-3252.
    IMAMVERDIYEV Y, SUKHOSTAT L. Lithological facies classification using deep convolutional neural network[J]. Journal of Petroleum Science and Engineering, 2019, 174: 216-228. doi: 10.1016/j.petrol.2018.11.023
    ZHANG Y, WANG G, LI M, et al. Automated classification analysis of geological structures based on images data and deep learning model[J]. Applied Sciences, 2018, 8(12): 2493. doi: 10.3390/app8122493
    PAN S J, YANG Q. A survey on transfer learning[J]. IEEE Transactions on Knowledge and Data Engineering, 2010, 22(10): 1345-1359. doi: 10.1109/TKDE.2009.191
    QURESHI A S, KHAN A, ZAMEER A, et al. Wind power prediction using deep neural network based meta regression and transfer learning[J]. Applied Soft Computing, 2017, 58: 742-755. doi: 10.1016/j.asoc.2017.05.031
    张野,李明超,韩帅. 基于岩石图像深度学习的岩性自动识别与分类方法[J]. 岩石学报,2018,34(2): 333-342.

    ZHANG Ye, LI Mingchao, HAN Shuai. Automatic identification and classification in lithology based on deep learning in rock images[J]. Acta Petrologica Sinica, 2018, 34(2): 333-342.
    LI N, HAO H, GU Q, et al. A transfer learning method for automatic identification of sandstone microscopic images[J]. Computers & Geosciences, 2017, 103: 111-121.
    WANG C, LI Y, FAN G, et al. Quick recognition of rock images for mobile applications[J]. Journal of Engineering Science and Technology Review, 2018, 11(4): 111-117. doi: 10.25103/jestr.114.14
    SANDLER M, HOWARD A, ZHU M, et al. MobileNet V2: inverted residuals and linear bottlenecks[C]//IEEE Conference on Computer Vision and Pattern Recognition. Salt Lake City: IEEE, 2018: 4510-4520.
    CHOLLET F. Xception: deep learning with depthwise separable convolutions[C]//IEEE Conference on Computer Vision and Pattern Recognition. Honolulu: IEEE, 2017: 1800-1807.
    PAN S J, TSANG I W, KWOL J T, et al. Domain adaptation via transfer component analysis[J]. IEEE Transactions on Neural Networks, 2011, 22(2): 199-210. doi: 10.1109/TNN.2010.2091281
    HE K, ZHANG X, REN S, et al. Deep residual learning for image recognition[C]//IEEE Conference on Computer Vision and Pattern Recognition. Las Vegas: IEEE, 2016: 770-778.
    杜恩宇,张宁,李艳荻. 基于自适应分块编码SVM的车道导向箭头多分类方法[J]. 光学学报,2018,38(10): 251-258.

    DU Enyu, ZHANG Ning, LI Yandi. Multi classification method of lane arrow markings based on support vector machines with adaptive partitioning coding[J]. Acta Optica Sinica, 2018, 38(10): 251-258.
  • 加载中
图(7) / 表(7)
计量
  • 文章访问数:  565
  • HTML全文浏览量:  277
  • PDF下载量:  21
  • 被引次数: 0
出版历程
  • 收稿日期:  2019-11-04
  • 修回日期:  2020-04-27
  • 网络出版日期:  2020-12-15
  • 刊出日期:  2021-04-15

目录

    /

    返回文章
    返回