Rock image lithology recognition method based on lightweight convolutional neural network
-
摘要:
岩性识别是油气勘探和开发过程中的重要环节，对于油气勘探定位、储层评价以及储层模型建立具有重要的指导意义。但传统的人工岩性识别方法耗时耗力，经典的深度学习模型虽然识别精度高，但模型的参数量较大，为了提高模型识别精度，同时降低模型的参数量，使模型适用于岩性实时识别工作，首先收集了白云岩、砂岩等8种岩石共
3016 张岩石图像构建岩性识别数据集，然后以轻量型卷积神经网络ShuffleNetV2模型为基础网络，提出了一种Rock-ShuffleNetV2岩性识别模型（RSHFNet模型）。模型中将混合注意力机制模块（convolutional block attention module，简称CBAM）以及多尺度特征融合模块（multi-scale feature fusion module，简称MSF）融入基础网络中以加强模型的特征提取能力，提升模型识别性能，并优化模型中ShuffleNetV2单元的堆叠次数以减少模型参数量。结果表明：与基础模型相比，RSHFNet模型的准确率达到了87.21%，提高了4.98%；同时，模型参数量与浮点运算量分别降低到了869702 个，0.93×108，分别是基础模型的0.67，0.63倍，模型参数量明显降低；并且RSHFNet模型的综合性能明显优于现有的卷积神经网络。RSHFNet岩性识别模型具有较高的识别精度和较好的泛化能力，同时更加的轻量化，为实现野外实时的岩性识别工作提供了新思路。-
关键词:
- 岩性识别 /
- ShuffleNetV2网络 /
- 混合注意力机制模块 /
- 多尺度特征融合模块 /
- 卷积神经网络
Abstract:ObjectiveLithology identification is a crucial step in the process of oil and gas detection and exploration, providing important guidance for exploration positioning, reservoir evaluation, and the establishment of reservoir models. However, traditional manual lithology identification methods are time-consuming and labor-intensive. Although classical deep learning models achieve high identification accuracy, they often have a large number of parameters. To enhance model accuracy while reducing the number of parameters, the aim of this research is to make the model suitable for real-time lithology identification.
MethodsThis paper first collected a dataset of
3016 rock images consisting of eight types of rocks, including dolomite and sandstone. Based on the lightweight convolutional neural network ShuffleNetV2, the paper proposes a Rock-ShuffleNetV2 lithology identification model (RSHFNet model). The model incorporates the convolutional block attention module (CBAM) and multi-scale feature fusion module (MSF) into the basic network to enhance feature extraction capabilities and improve identification performance. Additionally, the number of stacked ShuffleNetV2 units is optimized to reduce the model's parameters.ResultsThe experimental results show that the RSHFNet model achieved an accuracy of 87.21%, which is a 4.98% improvement over the baseline model. Furthermore, the model's parameters and floating-point operations were reduced to
869702 , 0.93×108, respectively, representing 67% of the model's parameters and 63% of the floating-point operations of the baseline model. This reduction significantly decreases the model's size. Additionally, the RSHFNet model demonstrates superior overall performance compared to existing convolutional neural networks.ConclusionThe proposed RSHFNet lithology identification model offers high recognition accuracy and strong generalization capabilities while being more lightweight, providing a new approach for real-time lithology identification in the field.
-
图 1 ShuffleNetV2核心结构
a. 通道混洗；b. 分组卷积；c. 下采样单元；d. 基本单元；BN. 只进行批量化处理；BN Relu. 先进行批量化处理，再用Relu激活函数处理；下同
Figure 1. Core structure of ShuffleNetV2
图 2 RSHFNet模型结构
MSF. 多尺度特征融合模块；CBAM. 混合注意力机制模块；下同
Figure 2. Structure of RSHFNet model
图 5 岩性识别数据集示例图像
a. 白云岩；b. 方解石；c. 菱锰矿；d. 菱镁矿；e. 石灰岩；f. 砂岩；g. 页岩；h. 泥岩
Figure 5. Example images of lithology identification dataset
图 6 数据增强后图像效果
a. 原始图像；b. 图像翻转；c. 图像旋转；d. 图像平移；e. 亮度调整；f. 色度调整；g. 对比度调整；h. 锐度调整；i. 裁剪拼接
Figure 6. Effects of data augmentation on images
图 7 改进前后混淆矩阵
a. ShuffleNetV2混淆矩阵；b. RSHFNet混淆矩阵；1. 白云岩；2. 方解石；3. 菱镁矿；4. 菱锰矿；5. 石灰岩；6. 砂岩；7. 泥岩；8. 页岩
Figure 7. Confusion matrix before and after improvement
图 8 RSHFNet模型预测结果
a～c. 预测类别：方解石；d～f. 预测类别：泥岩；g～i. 预测类别：白云岩；j～l，t. 预测类别：石灰岩；m～o，w. 预测类别：砂岩；p～r. 预测类别：菱锰矿；s，u. 预测类别：菱镁矿；v，x. 预测类别：页岩
Figure 8. Predicted results of the RSHFNet model
表 1 数据增强前后训练集分布
Table 1. Distributions of training set before and after data augmentation
岩性 原始训练集
数量/张增强后训练集
数量/张岩性 原始训练集
数量/张增强后训练集
数量/张白云岩 343 1 372 石灰岩 228 1 368 方解石 412 1 236 砂岩 495 1 485 菱镁矿 45 360 页岩 417 1 251 菱锰矿 60 466 泥岩 414 1 242 
下载: 导出CSV
表 2 数据增强前后实验结果
Table 2. Experimental results before and after data augmentation
数据集 准确率/% 精确率/% 召回率/% F1分数/% 原始数据集 76.58 69.93 68.85 69.10 数据增强后的数据集 82.23 82.21 81.42 81.54
下载: 导出CSV
表 3 引入不同注意力模块后的实验结果
Table 3. Experimental results after introducing different attention modules
模型 准确
率/%精确
率/%召回
率/%F1分
数/%模型参
数量/个浮点运
算量/108基础网络 82.23 82.21 81.42 81.54 1 295 037 1.47 基础网络+SE 82.89 81.30 80.21 80.19 1 303 064 1.48 基础网络+ECA 83.32 83.80 81.21 82.15 1 260 268 1.48 基础网络+CA 83.06 83.02 79.76 80.93 1 325 720 1.48 基础网络+CBAM 86.05 86.09 84.75 85.12 1 304 648 1.48
下载: 导出CSV
表 4 消融实验结果
Table 4. Results of ablation experiment
模型 准确
率/%精确
率/%召回
率/%F1分
数/%模型参
数量/个浮点运算
量/108基础网络 82.23 82.21 81.42 81.54 1 295 037 1.47 基础网络+CBAM 86.05 86.09 84.75 85.12 1 304 648 1.48 基础网络+MSF 83.39 82.84 83.31 82.90 1 260 404 1.51 基础网络-Lighter 85.71 87.41 85.89 83.38 850 848 0.90 RSHFNet 87.21 86.37 86.68 86.39 869 702 0.93
下载: 导出CSV
表 5 不同模型的对比实验结果
Table 5. Comparative experimental results of different models
模型 准确
率/%精确
率/%召回
率/%F1分
数/%模型参
数量/个浮点运算
量/108VGG16 67.77 70.68 67.66 66.96 134 301 768 155.20 ResNet18 85.71 84.86 86.29 85.27 11 180 616 18.24 ResNet50 85.05 85.09 83.73 84.11 23 524 424 41.32 DenseNet169 87.21 85.63 86.83 85.91 12 497 800 34.34 GhostNet 82.39 84.18 82.80 83.09 4 212 120 1.97 MobileNetV3-small 82.72 78.18 78.72 78.71 1 238 996 0.65 MobileNetV3-large 85.88 83.51 82.56 82.83 2 685 164 2.71 ShufleNetV2 82.23 82.21 81.42 81.54 1 295 037 1.47 ShuffleNetV2-Y 79.57 79.31 78.24 78.13 275 066 0.33 RSHFNet 87.21 86.37 86.68 86.39 869 702 0.93
下载: 导出CSV
-
[1] 陈丽, 刘娟, 王末, 等. 中国油料作物能源利用效率与温室气体排放[J]. 中国生态农业学报(中英文), 2023, 31(12): 1984-1996.CHEN L, LIU J, WANG M, et al. Energy use efficiency and greenhouse gas emissions of oil crops in China[J]. Chinese Journal of Eco-Agriculture, 2023, 31(12): 1984-1996. (in Chinese with English abstract [2] ZOU C N, YANG Z, ZHU R K, et al. Progress in China's unconventional oil & gas exploration and development and theoretical technologies[J]. Acta Geologica Sinica (English Edition), 2015, 89(3): 938-971. doi: 10.1111/1755-6724.12491 [3] 王婷婷, 孙振轩, 戴金龙, 等. 松辽盆地中央坳陷区储层岩性智能识别方法[J]. 吉林大学学报(地球科学版), 2023, 53(5): 1611-1622.WANG T T, SUN Z X, DAI J L, et al. Intelligent identification method of reservoir lithology in Central Depression of Songliao Basin[J]. Journal of Jilin University (Earth Science Edition), 2023, 53(5): 1611-1622. (in Chinese with English abstract [4] 胡前泽, 王玲利, 崔娥. 三塘湖盆地致密油储层评价及成藏主控因素[J]. 大庆石油地质与开发, 2013, 32(3): 164-169. doi: 10.3969/J.ISSN.1000-3754.2013.03.033HU Q Z, WANG L L, CUI E. Tight reservoir evaluation of the unconventional oil and study on the main controlling factors of hydrocarbon accumulation in Santanghu Basin[J]. Petroleum Geology & Oilfield Development in Daqing, 2013, 32(3): 164-169. (in Chinese with English abstract doi: 10.3969/J.ISSN.1000-3754.2013.03.033 [5] 陈国俊, 吕成福, 王琪, 等. 珠江口盆地深水区白云凹陷储层孔隙特征及影响因素[J]. 石油学报, 2010, 31(4): 566-572. doi: 10.7623/syxb201004008CHEN G J, LYU C F, WANG Q, et al. Characteristics of pore evolution and its controlling factors of Baiyun Sag in deepwater area of Pearl River Mouth Basin[J]. Acta Petrolei Sinica, 2010, 31(4): 566-572. (in Chinese with English abstract doi: 10.7623/syxb201004008 [6] MŁYNARCZUK M, GÓRSZCZYK A, ŚLIPEK B. The application of pattern recognition in the automatic classification of microscopic rock images[J]. Computers & Geosciences, 2013, 60: 126-133. [7] 付光明, 严加永, 张昆, 等. 岩性识别技术现状与进展[J]. 地球物理学进展, 2017, 32(1): 26-40. doi: 10.6038/pg20170104FU G M, YAN J Y, ZHANG K, et al. Current status and progress of lithology identification technology[J]. Progress in Geophysics, 2017, 32(1): 26-40. (in Chinese with English abstract doi: 10.6038/pg20170104 [8] MARTIN K G, CARR T R. Developing a quantitative mudrock calibration for a handheld energy dispersive X-ray fluorescence spectrometer[J]. Sedimentary Geology, 2020, 398: 105584. doi: 10.1016/j.sedgeo.2019.105584 [9] SINGH N, SINGH T N, TIWARY A, et al. Textural identification of basaltic rock mass using image processing and neural network[J]. Computational Geosciences, 2010, 14(2): 301-310. doi: 10.1007/s10596-009-9154-x [10] 郭超, 刘烨. 多色彩空间下的岩石图像识别研究[J]. 科学技术与工程, 2014, 14(18): 247-251. doi: 10.3969/j.issn.1671-1815.2014.18.048GUO C, LIU Y. Recognition of rock images based on multiple color spaces[J]. Science Technology and Engineering, 2014, 14(18): 247-251. (in Chinese with English abstract doi: 10.3969/j.issn.1671-1815.2014.18.048 [11] 程国建, 殷娟娟. 基于SVM的岩石薄片图像分类[J]. 科技创新与应用, 2015, 5(1): 38.CHENG G J, YIN J J. Classification of rock slice images based on SVM[J]. Technology Innovation and Application, 2015, 5(1): 38. (in Chinese with English abstract [12] DENG C X, PAN H P, FANG S N, et al. Support vector machine as an alternative method for lithology classification of crystalline rocks[J]. Journal of Geophysics and Engineering, 2017, 14(2): 341-349. doi: 10.1088/1742-2140/aa5b5b [13] 张驰, 潘懋, 胡水清, 等. 融合储层纵向信息的机器学习岩性识别方法[J]. 地质科技通报, 2023, 42(3): 289-299.ZHANG C, PAN M, HU S Q, et al. A machine learning lithologic identification method combined with vertical reservoir information[J]. Bulletin of Geological Science and Technology, 2023, 42(3): 289-299. (in Chinese with English abstract [14] SIMONYAN K, ZISSERMAN A. Very deep convolutional networks for large-scale image recognition[J]. Computer Science, 2014: 1409.1556. [15] HE K M, ZHANG X Y, REN S Q, et al. Deep residual learning for image recognition[C]//Anon. 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). [S.l.]: IEEE, 2016: 770-778. [16] MA N N, ZHANG X Y, ZHENG H T, et al. ShuffleNet V2: Practical guidelines for efficient CNN architecture design[M]// Computer vision-ECCV 2018. Cham: Springer International Publishing, 2018: 122-138. [17] 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]//Anon. 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). [S.l.]: IEEE, 2023: 7464-7475. [18] 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.2017.04.020CHENG G J, GUO W H, FAN P Z. 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. (in Chinese with English abstract doi: 10.3969/j.issn.1673-064X.2017.04.020 [19] 李娜, 顾庆, 姜枫, 等. 一种基于卷积神经网络的砂岩显微图像特征表示方法[J]. 软件学报, 2020, 31(11): 3621-3639.LI N, GU Q, JIANG F, et al. Feature representation method of microscopic sandstone images based on convolutional neural network[J]. Journal of Software, 2020, 31(11): 3621-3639. (in Chinese with English abstract [20] 谭永健, 田苗, 徐德馨, 等. 基于Xception网络的岩石图像分类识别研究[J]. 地理与地理信息科学, 2022, 38(3): 17-22. doi: 10.3969/j.issn.1672-0504.2022.03.003TAN Y J, TIAN M, XU D X, et al. Rock image classified identification based on Xception network[J]. Geography and Geo-Information Science, 2022, 38(3): 17-22. (in Chinese with English abstract doi: 10.3969/j.issn.1672-0504.2022.03.003 [21] RAN X J, XUE L F, ZHANG Y Y, et al. Rock classification from field image patches analyzed using a deep convolutional neural network[J]. Mathematics, 2019, 7(8): 755. doi: 10.3390/math7080755 [22] 马泽栋, 马雷, 李科, 等. 基于岩石图像深度学习的多尺度岩性识别[J]. 地质科技通报, 2022, 41(6): 316-322.MA Z D, MA L, LI K, et al. Multi-scale lithology recognition based on deep learning of rock images[J]. Bulletin of Geological Science and Technology, 2022, 41(6): 316-322. (in Chinese with English abstract [23] SZEGEDY C, LIU W, JIA Y Q, et al. Going deeper with convolutions[C]//Anon. 2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). [S.1.]: IEEE, 2015: 1-9. [24] 高广尚. 深度学习推荐模型中的注意力机制研究综述[J]. 计算机工程与应用, 2022, 58(9): 9-18. doi: 10.3778/j.issn.1002-8331.2112-0382GAO G S. Survey on attention mechanisms in deep learning recommendation models[J]. Computer Engineering and Applications, 2022, 58(9): 9-18. (in Chinese with English abstract doi: 10.3778/j.issn.1002-8331.2112-0382 [25] WOO S, PARK J, LEE J Y, et al. CBAM: Convolutional block attention module[M]//Computer vision-ECCV 2018. Cham: Springer International Publishing, 2018: 3-19. [26] HUANG G, SUN Y, LIU Z, et al. Deep networks with stochastic depth[M]//Computer vision-ECCV 2016. Cham: Springer International Publishing, 2016: 646-661. [27] 马永生, 蔡勋育, 赵培荣. 深层、超深层碳酸盐岩油气储层形成机理研究综述[J]. 地学前缘, 2011, 18(4): 181-192.MA Y S, CAI X Y, ZHAO P R. The research status and advances in porosity evolution and diagenesis of deep carbonate reservoir[J]. Earth Science Frontiers, 2011, 18(4): 181-192. (in Chinese with English abstract [28] 盛双占, 潘海峰, 石湘, 等. 吐哈盆地胜北凹陷三间房组致密砂岩气藏沉积特征及有利区优选[J]. 地质科技通报, 2024, 43(4): 181-190.SHENG S Z, PAN H F, SHI X, et al. Sedimentary characteristics and favourable area selection of tight sandstone gas reservoirs in the Sanjianfang Formation of the Shengbei Depression, Tuha Basin[J]. Bulletin of Geological Science and Technology, 2024, 43(4): 181-190. (in Chinese with English abstract [29] 付小平, 刘苗苗. 涪陵地区凉高山组富有机质泥岩微相特征及油气富集规律[J]. 断块油气田, 2023, 30(2): 230-237.FU X P, LIU M M. Microfacies characteristics of organic-rich mudstone and oil and gas enrichment law of Lianggaoshan Formation in Fuling area[J]. Fault-Block Oil & Gas Field, 2023, 30(2): 230-237. (in Chinese with English abstract [30] SHORTEN C, KHOSHGOFTAAR T M. A survey on image data augmentation for deep learning[J]. Journal of Big Data, 2019, 6(1): 60. doi: 10.1186/s40537-019-0197-0 [31] 袁硕, 刘玉敏, 安志伟, 等. 基于改进ShuffleNetV2网络的岩石图像识别[J]. 吉林大学学报(信息科学版), 2023, 41(3): 450-458. doi: 10.3969/j.issn.1671-5896.2023.03.009YUAN S, LIU Y M, AN Z W, et al. Rock image recognition based on improved ShuffleNetV2 network[J]. Journal of Jilin University (Information Science Edition), 2023, 41(3): 450-458. (in Chinese with English abstract doi: 10.3969/j.issn.1671-5896.2023.03.009 -
下载:
点击查看大图
计量
- 文章访问数: 1030
- PDF下载量: 164
- 被引次数: 0
