钻孔岩性柱状图是地质勘探和资源开发的主要成果之一，在过去，钻孔岩性柱状图主要通过手工绘制完成，存在劳动强度大、效率低下的问题。尽管后期使用Excel、AutoCAD以及相关编程语言提高了自动化程度，但仍然需要人工干预，受个人经验知识影响较大，达不到完全的自动分层与成图的效果。

随着智能识别技术的的快速发展，使得对钻探测井技术成果的岩性柱状智能识别与自动分层成为可能。基于此，本研究实现了基于计算机视觉的钻孔岩性柱状自动分层算法：首先对YCJ90/360(A)钻孔测井分析仪采集的同步视频数据，采用视频取帧、去除模糊帧、数据增广和半智能标注等方法构建了钻孔煤岩检测数据集；然后针对钻孔煤岩数据集中煤岩类间差别小和孔壁外观类目标小导致的检测精度低、效果不佳的问题，对YOLOv5的网络结构进行检测头分支结构、骨干网络注意力机制融合、锚框参数和边框回归损失函数改进；其后，对专门用于煤矿井下的综合测井技术——YCJ90/360(A)钻孔测井分析仪测井得到的自然伽马数据预处理和聚类分析，实现岩性自动分层；最后协同自然伽马测井曲线自动分层和随钻视频智能识别技术，设计了钻孔岩性柱状自动分层算法，并采用平均分层综合匹配度和平均岩性综合匹配度这两个指标评估了钻孔岩性柱状自动分层的准确性。研究结果表明：①构建的钻孔煤岩检测数据集共14类、45509个目标；②改进YOLOv5算法的平均精度均值、准确率和召回率相较于YOLOv5算法分别提升了9.86%、5.38%和4.96%；③钻孔岩性柱状自动分层平均分层综合匹配度为85.05%，平均岩性匹配度高达94.02%，验证了钻孔岩性柱状自动分层算法的准确性。

综上所述，本研究完成了基于计算机视觉的钻孔岩性柱状自动分层研究，并且采用本研究成果针对孟村矿6#、9#和10#超前钻探孔进行了实验分析，验证了本研究所提出的基于计算机视觉的钻孔岩性自动分层算法对与钻孔的分层层位和岩性判定的有效性与实用性，为测井数据的成果解释提供了一种智能化的钻孔岩性自动分层方法。

Drill hole lithology column chart is one of the main results of geological exploration and resource development. In the past, drill hole lithology column chart was mainly done by manual drawing, which had the problems of high labor intensity and low efficiency. Although automation has been improved by using Excel, AutoCAD and related programming languages in later years, it still requires manual intervention and is influenced by personal experience and knowledge, and cannot achieve the effect of fully automatic stratification and mapping.

With the rapid development of intelligent recognition technology, it makes it possible to intelligently identify and automatically stratify the lithological column of drilling and logging technology results. Based on this, this study implements a computer vision-based borehole lithology column automatic stratification algorithm: firstly, a borehole coal rock detection dataset is constructed using video frame extraction, blur frame removal, data broadening and semi-intelligent labeling for the synchronized video data collected by YCJ90/360(A) borehole logging analyzer; then, for the problems of low detection accuracy and poor results caused by small differences between coal rock classes and small targets of borehole wall appearance classes in the borehole coal rock dataset, the YOLOv5 is used to identify and automatically stratify the borehole coal rock. Then, the network structure of YOLOv5 is improved with the branching structure of detection head, the fusion of attention mechanism of backbone network, anchor frame parameters and loss function of border regression to address the problems of low detection accuracy and poor results caused by small differences in coal rock classes and small targets in hole wall appearance. The natural gamma data obtained from the logging of the analyzer were pre-processed and clustered to achieve automatic lithology stratification. Finally, the borehole lithology column automatic stratification algorithm was designed in collaboration with the natural gamma logging curve automatic stratification and the drilling video intelligent recognition technology, and the accuracy of the borehole lithology column automatic stratification was evaluated using two indexes, the average stratification comprehensive matching degree and the average lithology comprehensive matching degree. The results show that: (i) the constructed borehole coal rock detection dataset includes 14 categories and 45509 targets of coal, fine sandstone, medium sandstone, coarse sandstone, mudstone, sandy mudstone, carbonaceous mudstone, limestone, igneous rock, siltstone, magmatic rock, fracture development, annular fracture and pore wall fragmentation; (ii) the mean accuracy, accuracy and recall of the improved YOLOv5 algorithm compared with the YOLOv5 algorithm The average mean accuracy, accuracy and recall of the improved YOLOv5 algorithm improved by 9.86%, 5.38% and 4.96%, respectively, compared with the YOLOv5 algorithm; ③The average comprehensive stratification match of the borehole lithology column automatic stratification was 85.05%, and the average lithology match was as high as 94.02%, which verified the accuracy of the borehole lithology column automatic stratification algorithm.

In conclusion, this study has completed the study of automatic computer vision-based borehole lithology column stratification, and experimental analysis has been conducted using this research result for the over-drilled holes 6#, 9# and 10# of Mengcun Mine, which verifies the effectiveness and practicality of the computer vision-based borehole lithology automatic stratification algorithm proposed in this study for the stratification and lithology determination of boreholes, and provides an intelligent borehole lithology automatic stratification method for the interpretation of logging data results.

[1] 阎枫. 基于供应链协同的大型煤炭企业物资供应商选择模型构建及应用[D]. 太原: 太原理工大学, 2016.

[2] 王均才. 面向"危机管理"的煤矿安全生产管理模式研究——以枣矿集团高庄煤矿为例[D]. 天津: 天津大学, 2009.

[3] 杨睿. 煤炭作为主要能源的地位短期难改变[N]. 第一财经日报, 2015-01-13(A14).

[4] 王国法, 杜毅博, 徐亚军, et al. 中国煤炭开采技术及装备50年发展与创新实践——纪念《煤炭科学技术》创刊50周年[J]. 煤炭科学技术, 2023, 51(1): 1-18.

[5] 田瑞云. 中国煤矿安全生产标准化建设探讨[J]. 煤炭工程, 2018, 50(S1): 163-168.

[6] 张勇. 煤矿与非煤矿山安全生产形势对比分析研究[J]. 中国矿业, 2022, 31(12): 15-20..

[7] 王飞. 煤矿井下陷落柱探测技术研究[J]. 煤炭与化工, 2020, 43(10): 51-59.

[8] 孙照引. 地质构造及采空区探查钻探技术[J]. 中州煤炭, 2009, 161(5): 58-60.

[9] 张敬水. 测井技术在井下出水钻孔水源探查中的应用[J]. 江西煤炭科技, 2022, 175(3): 112-114.

[10] 宋梓豪. 煤矿井下钻孔雷达数据处理与解释方法研究[D]. 吉林: 吉林大学, 2022.

[11] 魏树群, 武剑. 综合钻孔测井技术在井下钻探工程的应用[J]. 河北化工, 2021, 44(7): 41-44.

[12] 乔杰, 王永文, 李智文, 等. 煤矿井下钻孔测井标准模型的构建研究[J]. 西部探矿工程, 2019, 31(11): 176-179.

[13] 王国法. 煤矿智能化最新技术进展与问题探讨[J]. 煤炭科学技术, 2022, 50(1): 1-27.

[14] 张金昌, 尹浩, 刘凡柏, 等. 自动化智能化地质岩芯钻探技术装备研发与应用[J]. 地质论评, 2022, 68(4): 1382-1392.

[15] 王国法. 《煤矿智能化建设指南（2021年版）》解读——从编写组视角进行解读[J]. 智能矿山, 2021, 2(4): 2-9.

[16] 王国法, 赵国瑞, 任怀伟. 智慧煤矿与智能化开采关键核心技术分析[J]. 煤炭学报, 2019, 44(1): 34-41.

[17] 石智军, 李泉新, 姚克. 煤矿井下智能化定向钻探发展路径与关键技术分析[J]. 煤炭学报, 2020, 45(6): 34-41.

[18] Peng F, Ya H, Song W, et al. Key Technology of Parameter Monitoring System for Directional Drilling Rig in Coal Mine[C]//Construction Industry Committee, China-Asia Economic Development Association. Conference Proceedings of the 8th International Symposium on Project Management. Australia: Aussino Academic Publishing House, 2020: 613-619.

[19] 袁亮, 张平松. 煤炭精准开采透明地质条件的重构与思考[J]. 煤炭学报, 2020, 45(7): 2346-2356.

[20] 王永文. 煤矿井下钻孔综合测井技术及应用[J]. 煤炭与化工, 2017, 40(5): 95-98.

[21] 许兴亮, 魏灏, 田素川, 等. 综放工作面煤柱尺寸对顶板破断结构及裂隙发育的影响规律[J]. 煤炭学报, 2015, 40(4): 850-855.

[22] 韩增强, 王川婴, 周济芳, 等. 基于钻孔图像的孔壁岩体完整性计算及在裂隙岩体灌浆效果评价中的应用[J]. 岩土工程学报, 2016, 38(S2): 245-249.

[23] 王国法. 煤矿智能化十大“痛点”解析及对策[J]. 智能矿山, 2021, 2(3): 1-4.

[24] Jegorowa Albina, Kurek Jarosław, Antoniuk Izabella, et al. Automatic Estimation of Drill Wear Based on Images of Holes Drilled in Melamine Faced Chipboard with Machine Learning Algorithms[J]. Forests, 2023, 14(2): 271-293.

[25] Sidorenko M, Orlov D, Ebadi M, et al. Deep learning in denoising of micro-computed tomography images of rock samples[J]. Computers & Geosciences, 2021, 151(2): 1-11.

[26] 袁祖贵. 采用自然伽马能谱测井仪研究天然放射性元素的高度和深度分布规律[J]. 原子能科学技术, 2005, 39(5): 466-469.

[27] Cripps A C, Mccann D M. The use of the natural gamma log in engineering geological investigations[J]. Engineering Geology, 2000, 55(4): 313-324.

[28] 肖波, 韩学辉, 周开金, 等. 测井曲线自动分层方法回顾与展望[J]. 地球物理学进展, 2010, 22(5): 1802-1810.

[29] 雷芬丽, 许平. 基于小波变换的测井多尺度分层研究[J]. 上海国土资源, 2015, 36(1): 76-79.

[30] 周锦程, 杨清亮, 张伟. 基于高斯小波变换的测井曲线自动分层模型[J]. 黔南民族师范学院学报, 2017, 37(4): 31-39.

[31] 张东晓, 陈云天, 孟晋. 基于循环神经网络的测井曲线生成方法[J]. 石油勘探与开发, 2018, 45(4): 598-607.

[32] Zhang L, Wang J, Jiao R L, et al. Delineation of bed boundaries of array induction logging curves using deep learning[J]. Applied Geophysics, 2021, 18(1): 45-53.

[33] L. Briqueu, N. Zaourar, C. Lauer-Leredde, et al. Wavelet based multiscale analysis of geophysical downhole measurements: Application to a clayey siliclastic sequence[J]. Journal of Petroleum Science and Engineering, 2009, 71(3): 112-120.

[34] R. Arabjamaloei, S. Edalatkha, E. Jamshidi, et al. Exact Lithologic Boundary Detection Based on Wavelet Transform Analysis and Real-Time Investigation of Facies Discontinuities Using Drilling Data[J]. Petroleum Science and Technology, 2011, 29(6): 569-578.

[35] Bappa Mukherjee, V.Srivardhan, P.N.S. Roy. Identification of formation interfaces by using wavelet and Fourier transforms[J]. Journal of Applied Geophysics, 2016, 128(16): 140-149.

[36] 司垒, 王忠宾, 熊祥祥, 等. 基于改进U-net网络模型的综采工作面煤岩识别方法[J]. 煤炭学报, 2021, 46(S01): 578-589.

[37] 田慧卿, 魏忠义. 基于图像识别技术的煤岩识别研究与实现[J]. 西安工程大学学报, 2012, 26(05): 657-660.

[38] 邹先坚, 王川婴, 韩增强, 等. 全景钻孔图像中结构面全自动识别方法研究[J]. 岩石力学与工程学报, 2017, 36(8): 1910-1920.

[39] 雷世威, 肖兴美, 张明. 基于改进YOLOv3的煤矸识别方法研究[J]. 矿业安全与环保, 2021, 48(3): 50-55.

[40] 王建才, 李瑾, 李志军, 等. 基于改进YOLOv5的煤岩图像检测识别研究[J]. 煤矿机械, 2022, 43(9): 204-208.

[41] Fariborz Goodarzi. The use of automated image analysis in coal petrology[J]. Canadian Journal of Earth Sciences, 1987, 24(5): 1064-1069.

[42] Ren J, Liu D. Recognition method of coal-rock interface based on time-domain vibration characteristics of coal cutter[J]. Coal Engineering, 2016, 48(03): 106-109.

[43] J. Busse, J.R. de Dreuzy, S. Galindo Torres, et al. Image processing based characterisation of coal cleat networks[J]. International Journal of Coal Geology, 2017, 169(2017): 1-21.

[44] Iwaszenko Sebastian, Róg Leokadia. Application of Deep Learning in Petrographic Coal Images Segmentation[J]. Minerals, 2021, 11(11): 1265.

[45] 王友林, 王建明, 程新星. 跨平台钻孔柱状图的自动生成[J]. 岩土工程技术, 2016, 030(1): 28-32.

[46] 范家爵. 用PC-1500袖珍计算机绘制钻孔柱状曲线图的方法[J]. 地质与勘探, 1986(09): 36-38.

[47] 张奔. 钻孔数据可视化方法研究[D]. 西安: 西安科技大学, 2014.

[48] 汪佳佳, 屠文森, 肖亮. 钻孔柱状图中的岩性花纹填充技术[J]. 科学技术与工程, 2014, 14(28): 242-246.

[49] 王友林, 王建明, 程新星. 跨平台钻孔柱状图的自动生成[J]. 岩土工程技术, 2016, 30(01): 28-32.

[50] 李志国, 王庶懋, 刘钦. 基于ArcGIS的钻孔柱状图自动生成方法[J]. 江西科学, 2022, 40(05): 990-996.

[51] Tilmann S E, Upchurch S B, Ryder G. Land Use Site Reconnaissance by Computer-Assisted Derivative Mapping[J]. Geological Society of America Bulletin, 1975, 86(1): 23-34.

[52] 杨建锋, 张翠光. 美国地质调查局环境地质科学现状、战略及启示[J]. 国土资源情报, 2013(10): 16-22.

[53] 李广辉, 徐彬, 雷焱, 等. 加拿大地质调查局2013～2018年战略新动向[J]. 国土资源情报, 2015(4): 45-51.

[54] Cripps A C, Mccann D M. The use of the natural gamma log in engineering geological investigations[J]. Engineering Geology, 2000, 55(4): 313-324.

[55] Day-Stirrat R J, Hillier S, Nikitin A, et al. Natural gamma-ray spectroscopy (NGS) as a proxy for the distribution of clay minerals and bitumen in the Cretaceous McMurray Formation, Alberta, Canada[J]. Fuel, 2021(Mar.15): 1-19.

[56] D'Andrade R G . Hierarchical Clustering[J]. Psychometrika, 2011, 43(1):59-67.

[57] Kaufman L, Rousseeuw P J. Agglomerative Nesting (Program AGNES)[M]//

Finding Groups in Data. New York: John Wiley & Sons, Ltd, 2008: 5.

[58] Kaufman L. Divisive Analysis (Program DIANA)[M]//Rousseeuw P J. Divisive Analysis (Program DIANA). Hoboken: John Wiley & Sons, Ltd, 2008: 2-6.

[59] Wong JAHA. Algorithm AS 136: A K-Means Clustering Algorithm[J]. Journal of the Royal Statistical Society, 1979, 28(1): 100-108.

[60] 张献民, 王建华. 工程数字测井自动分层的人工智能技术[J]. 天津大学学报: 自然科学与工程技术版, 2001, 34(5): 633-635.

[61] Oana Niculaescu. Classifying data with decision trees[J]. XRDS: Crossroads, The ACM Magazine for Students, 2018, 24(4): 55-57.

[62] 林香亮, 袁瑞, 孙玉秋, 等. 支持向量机的基本理论和研究进展[J]. 长江大学学报(自科版), 2018, 15(17): 48-53.

[63] Permanasari Y, Ruchjana B N, Hadi S, et al. Innovative Region Convolutional Neural Network Algorithm for Object Identification[J]. JOItmC, 2022, 8(4): 182.

[64] Girshick R. Fast R-CNN[C]//International Conference on Computer Vision. IEEE Computer Society, New York: IEEE, 2015: 1440-1448.

[65] Wu, ZhengminLuo, KunCao, et al. Fast location and classification of small targets using region segmentation and a convolutional neural network[J]. Computers and Electronics in Agriculture, 2020, 169(C): 105207-105215.

[66] Ren S, He K, Girshick R, et al. Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks[J]. IEEE Transactions on Pattern Analysis & Machine Intelligence, 2017, 39(6): 1137-1149.

[67] Redmon J, Divvala S, Girshick R, et al. You Only Look Once: Unified, Real-Time Object Detection[C]// ieee computer society. 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). New York: IEEE, 2016: 779-788.

[68] Bobkov A V, Aung Kh. Real-Time Person Identification by Video Image Based on YOLOv2 and VGG 16 Networks[J]. Automation and Remote Control, 2022, 83(10): 1567-1575.

[69] Hurtik Petr, Molek Vojtech, Hula Jan, et al. Poly-YOLO: higher speed, more precise detection and instance segmentation for YOLOv3[J]. Neural Computing and Applications, 2022, 34(10): 8275-8290.

[70] Roy A M, Bose R, Bhaduri J. A fast accurate fine-grain object detection model based on YOLOv4 deep neural network[J]. Neural Computing and Applications, 2022, 34(5): 3895-3921.

[71] Yanyang Zeng, Yang Yu, Zihan Zhou. Research on Small Target Recognition Model Based on Improved YOLOv5[J]. Journal of Research in Science and Engineering, 2022, 4(11): 28-33.

[72] Anamika Dhillon, Gyanendra K. Verma. Convolutional neural network: a review of models, methodologies and applications to object detection[J]. Progress in Artificial Intelligence, 2019, 9(prepublish): 82-112.

[73] Smith P I. Neural Networks[J]. Expert Systems, 2003, 24(12): 143-156.

[74] Narayan S. The Generalized Sigmoid Activation Function: Competitive Supervised Learning[J]. Information Sciences, 1997, 99(1-2): 69-82.

[75] 罗鹏, 李会方. 基于Tanh多层函数的量子神经网络算法及其应用的研究[J]. 计算机与数字工程, 2012, 40(01): 4-6.

[76] Fu X, Chen L, Wu F. ReLU artificial neural networks for the grid adaptation in finite element method[J]. Journal of Physics: Conference Series, 2021, 1978(1): 1-8.

[77] James G M. Variance and Bias for General Loss Functions[J]. Machine Learning, 2003, 51(2): 115-135.

[78] Hodson T O, Over T M, Foks S S. Mean squared error, deconstructed[J]. Journal of Advances in Modeling Earth Systems, 2021, 13(12): 1-10.

[79] Spurek P, Tabor J. Cross-Entropy Clustering[J]. Pattern Recognition, 2014, 47(9): 3046-3059.

[80] Rubio J, Angelov P, Pacheco J. Uniformly Stable Backpropagation Algorithm to Train a Feedforward Neural Network[J]. IEEE Transactions on Neural Networks, 2011, 22(3): 356-366.

[81] Bertels K, Neuberg L, Vassiliadis S, et al. On Chaos and Neural Networks: The Backpropagation Paradigm[J]. Artificial Intelligence Review, 2001, 15(3): 165-187.

[82] An M C, Lu H M, Hecht-Nielsen R. On the Geometry of Feedforward Neural Network Error Surfaces[J]. Neural Computation, 1993, 5(6): 910-927.

[83] Andrei Dmitri Gavrilov, Alex Jordache, Maya Vasdani, et al. Preventing Model Overfitting and Underfitting in Convolutional Neural Networks[J]. International Journal of Software Science and Computational Intelligence (IJSSCI), 2018, 10(4): 19-28.

[84] Ghosh R, Smadi O. Automated Detection and Classification of Pavement Distresses using 3D Pavement Surface Images and Deep Learning[J]. Transportation Research Record, 2021, 2675(9): 1359-1374.

[85] Valls Canudas Núria, Calvo Gómez Míriam, Golobardes Ribé Elisabet, et al. Use of Deep Learning to Improve the Computational Complexity of Reconstruction Algorithms in High Energy Physics[J]. Applied Sciences, 2021, 11(23): 11467-11471 .

[86] Lee C C, Rahiman M, Rahim R A, et al. A Deep Feedforward Neural Network Model for Image Prediction[J]. Journal of Physics: Conference Series, 2021, 1878(1): 1-5.

[87] Ververidis D, Kotropoulos C. Gaussian Mixture Modeling by Exploiting the Mahalanobis Distance[J]. IEEE Transactions on Signal Processing, 2008, 56(7): 2797-2811.