煤岩识别技术是实现煤矿智能化采掘的核心技术支撑和必由之路。然而，煤矿井下工作面复杂的环境条件对实现精确的煤岩识别提出了重大挑战。针对这一问题，本文围绕“采掘过程煤岩智能感知识别方法”这一关键科学问题展开研究。通过现场取样、室内试验、理论分析与仿真模拟的综合方法，以煤岩反射光谱、截割过程信号和视觉图像为研究对象，开展煤岩反射光谱实验、室内模拟截割试验以及煤岩图像数据采集工作，构建了煤岩物理信息数据库，分别提出了基于反射光谱、基于截割过程信号、基于视觉图像以及基于多传感器信息融合的煤岩识别技术。本文主要研究内容和成果如下：

（1）通过分析煤岩识别任务需求，确定了反射光谱、截割过程信号和视觉图像数据作为主要输入源。首先，基于煤岩反射光谱信息采集平台，建立了多状态、多参数组合的煤岩反射光谱数据库。针对典型煤岩样本的不同波段光谱特征进行了分析，讨论了煤岩物质成分对光谱特征的影响。其次，基于室内模拟煤岩截割试验平台，采集了截割过程中的扭矩、三向力和振动等数据，并建立了煤岩截割过程信号数据库。最后，利用煤矿井下工作面现场采集的煤岩图像数据，构建了煤岩图像数据库，并采用图像数据增强方法对数据进行了扩充。

（2）基于小波散射变换和双向长短期记忆网络，构建了煤岩反射光谱分类模型。利用小波散射网络有效提取了光谱特征，并将其输入至BiLSTM网络模型中进行训练和验证。结果显示，WST-BiLSTM模型在煤岩反射光谱分类识别上达到了99.4%的准确度，相较于其他网络模型，实现了更为精准的分类识别。

（3）基于连续小波变换和Alex Net网络，建立了煤岩截割过程信号分类模型。利用CWT将煤岩截割过程一维信号数据转化为二维时频图，并预加载Alex Net模型对煤岩截割过程信号时频图进行了训练和测试，结果表明，CWT-Alex Net模型取得了较高的分类准确度。

（4）利用深度可分离卷积和聚焦的线性注意力模块，改进了YOLOv8模型，基于迁移学习思想，构建了改进的YOLOv8煤岩图像分割模型。基于建立的煤岩视觉图像数据库，对该模型进行了训练和测试，并通过多项指标评估了模型的性能。结果显示，该模型对所有煤岩图像样本的预测精度和召回率达到了97.7%和99.7%。

（5）确定了以数据级和决策级融合为主要方向的煤岩识别融合策略。将反射光谱、截割过程信号和视觉图像数据进行融合分类，提高了识别的准确性，并提出了一种新的煤岩界面识别决策级融合方法，为智能化采掘提供了理论支撑和技术支持。

Coal rock identification technology is a core technical support and a necessary path for the realization of intelligent mining in coal mines. However, the complex environmental conditions of the coal mine working face pose significant challenges to achieving accurate coal rock identification. To address this issue, this paper focuses on the key scientific problem of "intelligent perception and identification methods for coal rock during the mining process." Through a comprehensive approach of field sampling, laboratory experiments, theoretical analysis, and simulation modeling, focusing on coal rock reflectance spectra, cutting process signals, and visual images as research subjects, this study conducted experiments on coal rock reflectance spectra, indoor simulated cutting tests, and coal rock image data collection. A physical information database of coal rock was constructed, and coal rock identification technologies based on reflectance spectra, cutting process signals, visual images, and the fusion of multisource heterogeneous data were proposed. The main research contents and results of this paper are as follows:

(1) By analyzing the requirements of the coal rock identification task, reflectance spectra, cutting process signals, and visual image data were determined as the main input sources. Firstly, a coal rock reflectance spectra database with multiple states and parameter combinations was established based on the coal rock reflectance spectra information collection platform. The spectral characteristics of different bands of typical coal rock samples were analyzed, and the impact of coal rock material composition on spectral characteristics was discussed. Secondly, based on the indoor simulated coal-rock cutting test platform, data such as torque, triaxial force, and vibration during the cutting process were collected, and a signal database for the coal-rock cutting process was established. Finally, using coal rock image data collected from the coal mine working face, a coal rock image database was constructed, and image data enhancement methods were applied to expand the data.

(2) Based on wavelet scattering transform and bidirectional long short-term memory networks, a coal rock reflectance spectra classification model was constructed. The wavelet scattering network effectively extracted spectral features, which were then fed into the BiLSTM network model for training and validation. The results showed that the WST-BiLSTM model achieved a 99.4% accuracy rate in coal rock reflectance spectra classification recognition, providing more precise classification recognition compared to other network models.

(3) Based on continuous wavelet transform and Alex Net network, a coal rock cutting process signal classification model was established. The CWT converted one-dimensional signal data of the coal rock cutting process into two-dimensional time-frequency maps, which were then trained and tested using a preloaded Alex Net model. The results indicated that the CWT-Alex Net model achieved high classification accuracy.

(4) By utilizing deep separable convolutions and focused linear attention modules, the YOLOv8 model was improved. Based on the transfer learning concept, an improved YOLOv8 coal rock image segmentation model was constructed. This model was trained and tested based on the established coal rock visual image database, and the model's performance was evaluated through various metrics. The results showed that the model achieved a prediction accuracy and recall rate of 97.7% and 99.7% for all coal rock image samples, respectively.

(5) A coal rock identification fusion strategy focusing on data-level and decision-level fusion was determined. The fusion classification of reflectance spectra, cutting process signals, and visual image data improved the accuracy of identification. A new decision-level fusion method for coal rock interface identification was proposed, providing theoretical support and technical assistance for intelligent mining.

[1] 王国法.加快煤矿智能化建设 推进煤炭行业高质量发展[J].中国煤炭,2021,47(01):2-10.

[2] 李洪雷.煤矿巷道掘锚一体自动化快速掘进关键技术研究[J]能源技术与管理,2019,44(01):133-134.

[3] 王国法,庞义辉,任怀伟等.智慧矿山系统工程及关键技术研究与实践[J/OL].煤炭学报:1-23.

[4] 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(01):1-27.

[5] 国家发展改革委、国家能源局应急部、国家煤矿安监局、工业和信息化部、财政部、科技部、教育部,关于加快煤智能化发展的指导意见[R]北京,2020.

[6] 赵学雷. 基于多传感器信息融合的载荷及煤岩判定与识别技术研究[D].中国矿业大学（北京）,2011.

[7] Wang J. Development and prospect on fully mechanized mining in Chinese coal mines[J]. International Journal of Coal Science and Technology. 2014,1(03):253-260.

[8] 张斌,苏学贵,段振雄等.YOLOv2在煤岩智能识别与定位中的应用研究[J].采矿与岩层控制工程学报,2020,2(02):94-101.

[9] 刘忠超,刘勇军.煤岩识别现状分析与发展方向[J].南阳理工学院学报,2018(4):26-30.

[10] Jay S, Maupas F, Bendoula R, et al. Retrieving LAI, chlorophyll and nitrogen contents in sugar beet crops from multi-angular optical remote sensing: Comparison of vegetation indices and PROSAIL inversion for field phenotyping[J]. Field crops research. 2017,210:33-46.

[11] Yang JD, Liu ZY, Yang Q. Simultaneous determination of chemical oxygen demand （COD） and biological oxygen demand （BOD5） in wastewater by near-infrared spectrometry. Journal of Water Resource and Protection[J]. 2009,1(04):286-289.

[12] Nicolai BM, Beullens K, Bobelyn E. Nondestructive measurement of fruit and vegetable quality by means of NIR spectroscopy: A review. Postharvest biology and technology[J]. 2007,46(02):99-118.

[13] Cloutis EA. Quantitative characterization of coal properties using bidirectional diffuse reflectance spectroscopy[J]. Fuel. 2003,82(18):2239-2254.

[14] Song Z, Kuenzer C. Spectral reflectance (400-2500nm) properties of coals, adjacent sediments, metamorphic and pyrometamorphic rocks in coal-fire areas: A case study of Wuda coalfield and its surrounding areas, northern China[J]. International Journal of Coal Geology. 2017,171:142-152.

[15] Jiang JY, Yang WH, Cheng YP, et al. Molecular structure characterization of middle-high rank coal via XRD, Raman and FTIR spectroscopy: Implications for coalification[J]. Fuel. 2019,239:559-572.

[16] Baldwin E, Han JL, Luo WT, et al. On fusion methods for knowledge discovery from multi-omics datasets[J]. Computational and Structural Biotechnology Journal,2020,18:509-517.

[17] 王水生.基于振动特性分析的采煤机煤岩识别控制系统[J].工矿自动化,2015,41(05):83-87.

[18] 李一鸣,白龙,蒋周翔等.基于EEMD-KPCA和KL散度的垮落煤岩识别[J].煤炭学报,2020,45(02):827-835.

[19] 尹玉玺,周常飞,许志鹏等.基于改进1DCNN的煤岩识别模型研究[J].工矿自动化,2023,49(01):116-122.

[20] 张强,张石磊,王海舰等.基于声发射信号的煤岩界面识别研究[J].电子测量与仪器学报,2017,31(02):230-237.

[21] Li Y, Cheng G, Chen X, et al. Coal–Rock Interface Recognition Based on Permutation Entropy of LMD and Supervised Kohonen Neural Network[J]. Current Science, 2019, 116(01).

[22] Ding ZW, Li XF, Huang X, et al. Feature extraction, recognition, and classification of acoustic emission waveform signal of coal rock sample under uniaxial compression[J]. International Journal of Rock Mechanics and Mining Science. 2022,160,105262.

[23] 付华,曹庆春.USMC控制的采煤机HHT-PCA-MRVM煤岩辨识算法[J].计算机应用与软件,2017,34(07):222-226+318.

[24] 王育龙.基于电流信号的煤岩识别方法研究[D].西安科技大学,2013.

[25] 程诚,刘送永.基于WPSV和BPNN的煤岩识别方法研究[J].煤炭工程,2018,50(01):108-112

[26] 田立勇,毛君,王启铭.基于采煤机摇臂惰轮轴受力分析的综合煤岩识别方法[J].煤炭学报,2016,41(03):782-787.

[27] 王元军,王明松,田山军等.基于卡尔曼滤波与随机森林的煤岩识别研究[J].煤炭技术,2021,40(12):208-211.

[28] 王海舰.煤岩界面多信息融合识别理论与实验研究[D].辽宁工程技术大学.

[29] 陈浜.基于视觉计算的煤岩识别方法研究[D].中国矿业大学(北京),2018.

[30] Bessinger SL, Nelson MG. Remnant roof coal thickness measurement with passive gamma ray instruments in coal mines[J]. IEEE Transactions on Industry Applications, 1993, 29(03):562-565.

[31] 王增才,富强.自然γ射线穿透煤层及支架顶梁衰减规律[J].辽宁工程技术大学学报,2006(06):6-9.

[32] 刘斌.综采面煤岩自然γ射线辐射规律及识别实验研究[D].中国矿业大学,2022.

[33] DANELS DJ. Short pulse radar for stratified lossy dielectric laver measurement[J].IEE Proceedings F-Communications, Radar and Signal Processing,1980,127(05):384-388.

[34] CHUFO RL, JOHNSON WJ. A radar coal thickness sensor[C]//IEEE Industry Application Society Annual Meeting, 1991:1182-1191.

[35] RALSTON JC, HAINSWORTH DW, MCPHEER RJ. Application of ground penetrating radar for coal thickness measurement[C], 1997,2:835-838.

[36] Strange AD, Chandran V, Ralston JC. Coal seam thickness estimation GPR and higher order statistics the near-surface case[C]//Proceedings of the Eighth International Symposium on. Australia, IEEE,2005:855-858.

[37] 刘帅,赵文生,高思伟.超宽带探地雷达煤层厚度探测试验研究[J].煤炭科学技术,2019,47(08):207-212.

[38] 许献磊,彭苏萍,马正等.基于空气耦合雷达的矿井煤岩界面随采动态探测原理及关键技术[J].煤炭学报,2022,47(08):2961-2977.

[39] Wang X, Ke X, et al. Characterization and Classification of Coals and Rocks Using Terahertz Time-Domain Spectroscopy[J]. Journal of Infrared, Millimeter, and Terahertz Waves, 2016, 38(02):1-13.

[40] 王昕.基于电磁波技术的煤岩识别方法研究[D].徐州:中国矿业大学,2017.

[41] 邵丹.基于太赫兹时域光谱技术和CA-PCA模型的煤岩分类方法研究[D].淮北师范大学,2022.

[42] 杨恩,王世博,葛世荣等.煤岩界面的高光谱识别原理[J].煤炭学报,2018,43(S2):646-653.

[43] Goetz A, Curtiss B, Shiley D A. Rapid gangue mineral concentration measurement over conveyors by NIR reflectance spectroscopy[C]// Elsevier Ltd. Elsevier Ltd, 2009:490-499.

[44] Taranik J V, Calvin W M, Kruse F A. Reflectance Spectroscopy Applied to Exploration for Mineral Deposits and Geothermal Systems, and to the Remediation of Mined Lands in the Great Basin of the Western United States.[C]// Art, Science and Applications of Reflectance Spectroscopy Symposium, 2010: 1-38.

[45] Song Z, KUNENZER C. Spectral reflectance (400–2500 nm) properties of coals, adjacent sediments, metamorphic and pyrometamorphic rocks in coal-fire areas: A case study of Wuda coalfield and its surrounding areas, northern China. [J].International Journal of Coal Geology.2017,171:142-152.

[46] 杨恩,王世博,葛世荣.典型煤系岩石的可见-近红外光谱特征研究[J].工矿自动化,2019,45(03):45-51+89.

[47] 杨恩,王世博,王赛亚等.典型煤岩反射光谱无监督感知方法研究[J].工矿自动化,2020,46(01):50-58.

[48] 吕渊博,王世博,葛世荣等.近红外光谱煤岩识别装置研制[J].工矿自动化,2022,48(07):32-42.

[49] 向阳,王世博,葛世荣等.粉尘环境下典型煤岩近红外光谱特征及识别方法[J].光谱学与光谱分析,2020,40(11):3430-3437.

[50] 张旭辉,张楷鑫,张超等.基于CARS与PCA的高光谱煤岩特征信息检测方法[J].西安科技大学学报,2020,40(05):760-768.

[51] 徐良骥,孟雪莹,韦任等.基于可见光-近红外光谱的煤岩识别方法实验研究[J].光谱学与光谱分析,2022,42(07):2135-2142.

[52] 韦任,徐良骥,孟雪莹等.基于高光谱特征吸收峰的煤岩识别方法[J].光谱学与光谱分析,2021,41(06):1942-1948.

[53] 孟雪莹.基于可见光-近红外光谱的煤岩识别方法试验研究[D].安徽理工大学,2020.

[54] 孙继平,陈浜.基于双树复小波域统计建模的煤岩识别方法[J].煤炭学报,2016,41(07):1847-1858.

[55] 孙继平,陈浜.基于小波域非对称广义高斯模型的煤岩识别算法[J].煤炭学报,2015,40(S2):568-575.

[56] 孙继平,陈浜.基于CLBP和支持向量诱导字典学习的煤岩识别方法[J].煤炭学报,2017,42(12):3338-3348.

[57] 伍云霞,张宏.基于Curvelet变换和压缩感知的煤岩识别方法[J].煤炭学报,2017,42(05):1331-1338.

[58] 伍云霞,孟祥龙.局部约束的自学习煤岩识别方法[J].煤炭学报,2018,43(09):2639-2646.

[59] 章华,李振璧,姜媛媛.基于图像纹理的煤岩识别研究[J].煤炭技术,2015,34(7):120-121.

[60] Liu X, Jing W, Zhou M, et al. Multi-Scale Feature Fusion for Coal-Rock Recognition Based on Completed Local Binary Pattern and Convolution Neural Network [J]. Entropy, 2019, 21(6):622.

[61] 王超,张强.基于LBP和GLCM的煤岩图像特征提取与识别方法[J].煤矿安全,2020,51(04):129-132.

[62] 韩晓天.基于纹理特征融合的煤岩图像分类方法研究[D].辽宁工程技术大学,2023.

[63] 张婷.基于变换域与高斯混合模型聚类的煤岩识别方法[J].煤炭技术,2018,37(11):320-323.

[64] 佘杰.基于图像的煤岩识别方法研究[D].北京:中国矿业大学（北京）,2014.

[65] 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(S1):578-589.

[66] 张斌,苏学贵,段振雄等.YOLOv2在煤岩智能识别与定位中的应用研究[J].采矿与岩层控制工程学报,2020,2(02):94-101.

[67] 孙传猛,王燕平,王冲等.融合改进YOLOv3与三次样条插值的煤岩界面识别方法[J].采矿与岩层控制工程学报,2022,4(01):81-90.

[68] 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(09):204-208.

[69] 李彦明.基于钻孔返渣图像的煤岩界面识别方法研究[J].煤矿安全,2021,52(03):175-179.

[70] Dong L H, Zhao P B. Application of Improved Canny Edge Detection Algorithm in Coal-Rock Interface Recognition [J]. Applied Mechanics and Materials, 2012, 220-223: 1279-1283.

[71] 黄韶杰,刘建功.基于高斯混合聚类的煤岩识别技术研究[J].煤炭学报,2015,40(S2):576-582.

[72] 吴德忠,刘泉声,黄兴等.基于边界跟踪和神经网络的煤岩界面识别方法研究[J].煤炭工程,2021,53(6):140-146.

[73] 高峰,殷欣,刘泉声等.基于塔式池化架构的采掘工作面煤岩图像识别方法[J].煤炭学报,2021,46(12):4088-4102.

[74] 张云,童亮,来兴平等.基于机器视觉的煤尘环境下掘进空间煤岩界面感知与精准识别[J].煤炭学报,2023:1-14.

[75] 闫志蕊,王宏伟,耿毅德.基于改进DeeplabV3+和迁移学习的煤岩界面图像识别方法[J].煤炭科学技术,2023,51(S1):429-439.

[76] Ren F, Liu ZY, Yang ZJ. Weighted Algorithm of Multi-Sensor Data Conflict in Coal-Rock Interface Recognition[J]. Applied Mechanics and Materials, 2011, 58-60:1908-1913.

[77] Wang HJ, Zhang Q. Dynamic Identification of Coal-Rock Interface Based on Adaptive Weight Optimization and Multi-Sensor Information Fusion[J]. Information Fusion, 2019;51:114-128.

[78] Liu Y, Dhakal S, Hao B. Coal and rock interface identification based on wavelet packet decomposition and fuzzy neural network[J]. Journal of Intelligent and Fuzzy Systems, 2020, 38(4):3949-3959.

[79] 梁义维,熊诗波.基于神经网络和Dempster-Shafter信息融合的煤岩界面预测[J].煤炭学报,2003(01):86-90.

[80] 杨健健,符世琛,姜海等.基于模糊判据的煤岩性状截割硬度识别[J].煤炭学报,2015,40(S2):540-545.

[81] 张强,王海舰,井旺等.基于模糊神经网络信息融合的采煤机煤岩识别系统[J].中国机械工程,2016,27(02):201-208.

[82] 雷静,余斌.基于信息融合和神经网络的煤岩识别方法[J].工矿自动化,2017,43(09):102-105.

[83] Si L, Wang ZB, Jiang G. Fusion Recognition of Shearer Coal-Rock Cutting State Based on Improved RBF Neural Network and D-S Evidence Theory[J]. IEEE Access, 2019.

[84] 司垒,王忠宾,李嘉豪等.基于图像和激光点云融合的智能采面煤岩识别[J].振动.测试与诊断,2023,43(02):254-262+407.

[85] Mallat S. Group invariant scattering. Communications on Pure and Applied Mathematics. 2012;65(10):1331-1398.

[86] Andén J, Mallat S. Deep scattering spectrum. IEEE Transactions on Signal Processing. 2014;62(16):4114-4128.

[87] Andén J, Lostanlen V, Mallat S. Joint time-frequency scat-tering for audio classification. IEEE International Workshop on Machine Learning for Signal Processing(MLSP)[C]. Boston,2015:1-6.

[88] Srinivasu PN, SivaSai JG, Ijaz MF, et al. Classification of skin disease using deep learning neural networks with MobileNet V2 and LSTM. Sensors. 2021;21(8):2852.

[89] Krishna SL, Jeya IJS, Deepa SN. Fuzzy-twin proximal SVM kernel-based deep learning neural network model for hyperspectral image classification. Neural Computing & Applications. 2022;34(21):19343-19376.

[90] Daubechies I. The wavelet transform, time-frequency localization and signal analysis. IEEE Transactions on Information Theory. 1990;36(5):961-1005.

[91] Fan X, Chen JY, Wang YH, et al. Intelligent recognition of coal mine microseismic signal based on wavelet scattering decomposition transform. Journal of China Coal Society. 2022;47(07):2722-2731.

[92] Bruna J and Mallat S. Invariant scattering convolution networks. IEEE Transactions on Pattern Analysis & Machine Intelligence. 2013;35(8):1872-1886.

[93] Hochreiter S, Schmidhuber J. Long short-term memory. Neural Computation. 1997;9(8):1735-1780.

[94] Graves A, Schmidhuber J. Framewise phoneme classification with bidirectional LSTM and other neural network architectures. Neural Network. 2005;18(5/6): 602-610.