1 School of Computer Science and Technology,Henan Polytechnic University,Henan Jiaozuo 454003,China
2 Henan Key Laboratory of Coal Measure Unconventional Resources Accumulation and Exploitation, School of Resources and Environment, Henan Polytechnic University, Henan Jiaozuo 454003, China
NIU Yongbin, born in 1980,is a professor at School of Resources and Environment,Henan Polytechnic University. He is mainly engaged in applied ichnology and sedimentolgy. E-mail: niuyongbin@hpu.edu.cn.
LU Bibo,born in 1978,is a professor at the School of Computer Science and Technology,Henan Polytechnic University. He is mainly engaged in artificial intelligence and image processing. E-mail: lubibo@hpu.edu.cn.
Bioturbation refers to various sedimentary textures or structures formed on sediment surfaces or within sediments due to biological activity. It plays a crucial role in analyzing paleoenvironmental conditions in sedimentary strata,predicting distribution patterns,evaluating the hydrocarbon generation capacity of source rocks,assessing the sealing capacity of caprocks,and revealing the mechanisms and effects of bioturbation on hydrocarbon reservoirs. Traditional methods for analyzing bioturbation intensity mainly relies on manual identification,followed by semi-quantitative classification using bioturbation index charts. This approach is highly subjective,inefficient,and prone to large errors. In this paper,we proposed a residual network model that incorporates an attention mechanism(Res-EMANet)by integrating the Efficient Multi-Scale Attention(EMA)mechanism into the ResNet-50 model. During training,the model employs stochastic gradient descent(SGD)with an initial learning rate of 0.01,a weight decay parameter of 0.0001,a batch size of 16,and a total of 300 epochs. Model performance improvements are evaluated based on five aspects: accuracy,precision,recall,F1-score,and the confusion matrix. We validated the model using a dataset of 3,028 core images from 16 wells of the Ordovician in the Tarim Basin,which contain various levels of bioturbation. The results show that: (1)The model can accurately classify bioturbation intensities ranging from level 0 to 5 in digital core images,achieving an accuracy of up to 91%. This significantly outperforms traditional manual methods as well as the original ResNet-50 model. (2)The model not only improves the accuracy of bioturbation grade recognition but also effectively reduces dependence on expert knowledge,as well as the labor intensity and subjectivity associated with manual bioturbation assessments. It demonstrates significant advantages in the automation,intelligence,and quantification of bioturbation feature analysis. This research offers an efficient and reliable quantitative analysis tool for the automated processing of bioturbation degree assessment and identification,which is of great significance to the sedimentology and paleontology studies in the field of oil and gas exploration.
LU Bibo,born in 1978,is a professor at the School of Computer Science and Technology,Henan Polytechnic University. He is mainly engaged in artificial intelligence and image processing. E-mail: lubibo@hpu.edu.cn.
"}, bioImg=null, bioContent=
LU Bibo,born in 1978,is a professor at the School of Computer Science and Technology,Henan Polytechnic University. He is mainly engaged in artificial intelligence and image processing. E-mail: lubibo@hpu.edu.cn.
在地质学研究中,生物在生命活动过程中对周围沉积物颗粒所进行的搅动、混合和破坏而形成的各种沉积结构或沉积构造被称为生物扰动(Hülse et al., 2022)。生物扰动会破坏和蚀变原生沉积的结构或构造,可增强或减弱油气水储集层质量及其流动特性(Taylor et al., 2003)。为了表征原始沉积地层被生物改造的程度,地质学家会评估沉积物受生物扰动的程度从而得到生物扰动指数(Bioturbation Index,BI),用于解释沉积环境和识别关键地层表面,确定源岩潜力、储集层质量和石油系统建模(Miller and Smail, 1997;Dey ans Sen, 2017)。因此,生物扰动的定量表征对重建生物生存的(古)环境和(古)生态以及研究生物扰动对储集层的改造效应具有重要意义(牛永斌等,2017)。
在油气、含水层和深海钻探计划的勘探过程中,钻井取得的岩心是地下生物遗迹和沉积学信息的宝贵来源(Sterling,2011)。文中选用新疆塔里木盆地塔河油田的奥陶系碳酸盐岩岩心作为研究对象。塔河油田面积近2400 km2,位于新疆维吾尔自治区塔里木盆地北缘,地处轮台县和库车县境内(张抗,1999)。塔河油田是中国发现的第1个陆上海相古生界超亿吨级大油田,奥陶系碳酸盐岩被认为是该地区主要的含油气层系之一(鲁新便等,2015;金强等,2020)。塔河油田奥陶系碳酸盐岩生物扰动强烈,具有很好的储集性能,是潜在的高质量油气储集层(艾合买提江等,2010;赵佳如等,2021;Eltom et al., 2023)。
塔河油田奥陶系岩心中生物扰动区域普遍发生白云石化作用,该成岩过程改变了储集物性特征(刘大卫等,2025),具体表现为白云石化作用提高了岩石的孔隙度和渗透率,使得生物扰动区域更易成为油气充填的对象,颜色较深(Niu et al., 2022)。即使生物扰动区域未经油气充填,但其矿物成分中白云石的含量较高,岩心在地表风化后颜色也较围岩基质更深。因此,岩心上生物扰动区域与围岩基质常呈现显著色差,且该特征在油气充注条件下更为显著。基于此,数字图像分析显示生物扰动区域多对应0~255灰度谱系中较低像素值的色阶范围(牛永斌等,2017)。
ResNet-50模型在减少模型参数的同时,又可以保持网络的高效表征能力,有效缓解了深度学习中常见的梯度消失问题(He et al., 2016)。近年来,ResNet被广泛应用于地质领域,在预测分类方面具有精度高的优点(刘大锰等,2024;刘艳如等,2025)。然而,其在处理塔河岩心图像中生物扰动复杂特征提取时仍然存在局限性。为了解决塔河岩心图像中的生物扰动特征在不同井下岩心上的表现不一及特征提取困难,笔者通过在ResNet-50模型上引入EMA注意力机制,提出Res-EMANet模型。
表 2中的结果表明,ResNet-50网络基础模型的各项指标相对较低,而通过引入了不同注意力机制后,模型的整体性能都有所提升。特别是,整合了EMA注意力机制的Res-EMANet模型在各项性能指标上均超越了使用CA(Channel Attention)、CBAM(Convolutional Block Attention Module)、SE(Squeeze-and-Excitation)和ECA(Efficient Channel Attention)注意力机制的模型(Hu et al., 2018;Woo et al., 2018;Wang et al., 2020;Hou et al., 2021)。
实验选取Res-EMANet模型,和其他当下流行的多种图像分类模型进行实验对比,以验证该模型的有效性,对比模型包括ResNet-50、Vgg-19、DenseNet121和EfficientNet-B0(Simonyan and Zisserman,2014;Huang et al., 2017;Tan and Le, 2019)。分别使用上述模型对文中的塔河岩心生物扰动数据集进行训练,记录每个训练周期的训练集准确率和测试准确率,以便及时掌握模型的训练情况,确保各个模型在收敛的状态下完成训练。训练过程可视化如图 7所示,其展示了5种卷积神经网络模型在训练集和验证集上的准确率曲线。
[AhmatjanA, ZhongJ H, LiY, ChenX. 2010. Stylolite characteristics and petroleum geology significance of Ordovician carbonate rocks in Tahe Oilfield. Journal of China University of Petroleum(Edition of Natural Science), 34(1): 7-11,17 ]
[JinQ, ZhangS, SunJ F, WeiH H, ChengF Q, ZhangX D. 2020. Formation and evolution of karst facies of Ordovician carbonate in Tahe Oilfield. Acta Petrolei Sinica, 41(5): 513-525 ]
[LiuD M, WangZ H, ChenJ M, QiuF, ZhuK, GaoL J, ZhouK Y, XuS B, SunF R. 2024. Classification of macerals and microfractures in deep coal seams based on ResNet: a case study of the No. 8 coal seam of the Carboniferous Benxi Formation in the Ordos Basin. Oil & Gas Geology, 45(6): 1524-1536 ]
[LiuD W, LiY T, HanJ, ZhangJ B, RuZ X, YangX Q, WangS, HuangC, XiaoC Y. 2025. Ultra-deep dolomite types and their reservoirs potential of the Ordovician Yingshan Formation in Shunbei area,Tarim Basin. Journal of Palaeogeography(Chinese Edition), 27(1): 126-140 ]
[LiuY R, WuX H, HeX H, LuoB B, TengQ Z. 2025. Research on core image classification based on improved ResNet50. Intelligent Com-puter and Applications, 15(2): 10-16 ]
[LuX B, HuW G, WangY, LiX H, LiT, LüY P, HeX M, YangD B. 2015. Characteristics and development practice of fault-karst carbonate reservoirs in Tahe area,Tarim Basin. Oil & Gas Geology, 36(3): 347-355 ]
[NiuY B, CuiS L, HuY Z, ZhongJ H, WangP J. 2017. Quantitative characterization of bioturbation based on digital image analysis of the Ordovician core from Tahe Oilfield of Tarim Basin. Journal of Palaeogeography(Chinese Edition), 19(2): 353-363 ]
[NiuY B, CuiS L, HuY Z, ZhongJ H, PanJ N. 2018. Three-dimensional reconstruction and their significance of bioturbation-type reservoirs of the Ordovician in Tahe Oilfield. Journal of Palaeogeography(Chinese Edition), 20(4): 691-702 ]
[NiuY B, XuZ L, LiuS X, ZhongJ H, ZhaoJ R, WangP J. 2020. Digital characterization and connectivity analysis of microcosmic pore structures of the Ordovician bioturbated carbonate rock reservoirs in Tahe Oilfield. Journal of Palaeogeography(Chinese Edition), 22(4): 785-798 ]
[NiuY B, ZhaoJ R, ZhongJ H, WangM, XuZ L, ChengM Y. 2021. Identification of the bioturbated carbonate reservoir and their porosity prediction based on conventional well logging data using artificial neural networks: take the Ordovician bioturbated carbonate reservoir in Tahe oilfield,Tarim Basin,as an example. Geological Review, 67(6): 1898-1909 ]
[NiuY B, JingC H, ShaoW M, ChengY G, LiZ Y. 2023. A review and perspective of bioturbated hydrocarbon and water reservoirs. Acta Sedimentologica Sinica, 41(6): 1934-1953 ]
[25]
张抗. 1999. 塔河油田的发现及其地质意义. 石油与天然气地质, 20(2): 24-28.
[26]
[ZhangK. 1999. The discovery of Tahe oilfield and its geological significance. Oil & Gas Geology, 20(2): 24-28 ]
[ZhouC Y, LiuW, WuT R, LiA, HanX S. 2024. Classification of rock thin section images based on mixture of expert models. Journal of Jilin University(Science Edition), 62(4): 905-914 ]
[ZhouY Z, ZuoR G, LiuG, YuanF, MaoX C, GuoY J, XiaoF, LiaoJ, LiuY P. 2021. The great-leap-forward development of mathematical geoscience during 2010-2019: big data and artificial intelligence algorithm are changing mathematical geoscience. Bulletin of Mineralogy,Petrology and Geochemistry, 40(3): 556-573 ]
[33]
AyranciK, YildirimI E, WaheedU B, MacEachernJ A. 2021. Deep learning applications in geosciences: insights into ichnological analysis. Applied Sciences, 11(16): 7736.
[34]
DeyJ, SenS. 2017. Impact of bioturbation on reservoir quality and production: a review. Journal of the Geological Society of India, 89(4): 460-470.
[35]
DoradorJ, Rodríguez-TovarF J, ExpeditionI. 2014. Quantitative estimation of bioturbation based on digital image analysis. Marine Geology,349: 55-60.
[36]
EltomH A, SyahputraM R N, El-HusseinyA, La CroixA D. . 2023. Spatial complexity of burrow attributes and their impact on porosity and permeability distributions in bioturbated reservoirs. Sedimentary Geology,450: 106395.
[37]
HeK M, ZhangX Y, RenS Q, SunJ. 2016. Deep residual learning for image recognition. 2016 IEEE Conference on Computer Vision and Pattern Recognition(CVPR). June 27-30,2016,Las Vegas,NV,USA. IEEE: 770-778.
[38]
HouQ B, ZhouD Q, FengJ S. 2021. Coordinate attention for efficient mobile network design. 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition(CVPR). June 20-25,2021. Nashville,TN,USA. IEEE: 13713-13722.
[39]
HuJ, ShenL, SunG. 2018. Squeeze-and-excitation networks. 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition. June 18-23,2018. Salt Lake City,UT,USA. IEEE: 7132-7141.
[40]
HuangG, LiuZ, VanDer Maaten L, WeinbergerK Q. 2017. Densely connected convolutional networks. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition: 4700-4708.
[41]
HülseD, VervoortP, van de VeldeS J, KanzakiY, BoudreauB, ArndtS, BottjerD J, HoogakkerB, KudererM, MiddelburgJ J, VolkenbornN, KirtlandTurner S, RidgwellA. 2022. Assessing the impact of bioturbation on sedimentary isotopic records through numerical models. Earth-Science Reviews,234: 104213.
[42]
KikuchiK, NaruseH. 2024. Abundance of trace fossil phycosiphon incertum in core sections measured using a convolutional neural network. Sedimentary Geology,461: 106570.
[43]
KnaustD. 2012. Methodology and Techniques. Developments in Sedimentology. Elsevier: 245-271.
[44]
KnaustD. 2017. Atlas of Trace Fossils in Well Core: Appearance, Taxonomy and Interpretation. Springer: 1-209.
[45]
Miguez-SalasO, DoradorJ, Rodríguez-TovarF J. 2019. Introducing Fiji and ICY image processing techniques in ichnological research as a tool for sedimentary basin analysis. Marine Geology,413: 1-9.
[46]
MillerM F, SmailS E. 1997. A semiquantitative field method for evaluating bioturbation on bedding planes. Palaios, 12(4): 391-396.
[47]
NiuY B, ChengM Y, ZhangL J, ZhongJ H, LiuS X, WeiD, XuZ L, WangP J. 2022. Bioturbation enhanced petrophysical properties in the Ordovician carbonate reservoir of the Tahe Oilfield,Tarim Basin,NW China. Journal of Palaeogeography, 11(1): 31-51.
[48]
OuyangD L, HeS, ZhangG Z, LuoM Z, GuoH Y, ZhanJ, HuangZ J. 2023. Efficient multi-scale attention module with cross-spatial learning. ICASSP 2023-2023 IEEE International Conference on Acoustics,Speech and Signal Processing: 1-5.
[49]
SimonyanK, ZissermanA. 2014. Very deep convolutional networks for large-scale image recognition. Arxiv Preprint: 1409.1556(publishied as a coference paper at ICLR 2015).
[50]
SterlingS N. 2011. Cores and core logging for geoscientists. Environmental and Engineering Geoscience,17: 307-308.
[51]
TanM, LeQ. 2019. Efficientnet: rethinking model scaling for convolutional neural networks. International conference on machine learning. PMLR: 6105-6114.
[52]
TaylorA M, GoldringR. 1993. Description and analysis of bioturbation and ichnofabric. Journal of the Geological Society, 150(1): 141-148.
[53]
TaylorA, GoldringR, GowlandS. 2003. Analysis and application of ichnofabrics. Earth-Science Reviews, 60(3-4): 227-259.
[54]
TimmerE R, GingrasM K, ZonneveldJ P. 2016. Pychno: a core-image quantitative ichnology logging software. Palaios, 31(11): 525-532.
[55]
TimmerE, KnudsonC, GingrasM. 2021. Applying deep learning for identifying bioturbation from core photographs. AAPG Bulletin, 105(4): 631-638.
[56]
WangQ L, WuB G, ZhuP F, LiP H, ZuoW M, HuQ H. 2020. Eca-Net: efficient channel attention for deep convolutional neural networks. 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition(CVPR). June 13-19,2020. Seattle,WA,USA. IEEE: 11534-11542.
[57]
WooS, ParkJ, LeeJ Y, KweonI S. 2018. Cbam: convolutional block attention module. Proceedings of the European Conference on Computer vision(ECCV): 3-19.
[58]
ReineckH E. 1963. Sedimentgefüge im Bereich der südlichen Nordsee. Abhandlungen der se Ckenbergischen Naturforschenden Gesellschaft,505: 1-138.