煤矸分选是煤炭生产中的重要环节，基于图像的煤矸分选技术因其显著的优点受到业界的广泛关注。在恶劣的煤炭生产环境下，高质量的图像难以获取，加之煤和矸石在物理性质上十分相似，识别困难。多光谱成像技术能够同时获取目标对象的图像和光谱信息，可提供更加丰富的数据信息。本文将该技术应用于煤矸识别，研究煤和矸石多光谱图像和光谱特征随照度、波段变化的规律，波段组合选择方法，识别方法等。具体工作如下：

（1）照度和波段对煤矸图像与光谱特征影响分析。设计不同光照强度、光谱波段下煤矸多光谱图像和光谱信息数据采集的实验方案。搭建实验平台，建立2个矿区5种不同光照强度、9个光谱波段下的煤和矸石的多光谱图像和光谱值数据集；分析煤和矸石图像特征标准差、偏度、熵、对比度，光谱特征共5个参数随光照强度、光谱波段的变化规律；对特征参数进行归一化处理，对比煤和矸石各参数偏差值受光照强度、光谱波段变化的波动情况。得到：煤和矸石多光谱图像特征和光谱特征均受光照强度和光谱波段变化的影响，并呈现不同的变化规律；煤和矸石多光谱图像特征和光谱特征偏差在7000lux照度下和585nm波段下较大，其他照度下较小。

（2）基于光谱波段组合选择的数据降维方法研究。在图像降维方面，分别采用主成分分析法（PCA）和基于类间特征可分性的最佳指数方法（OIF）对多光谱图像进行特征波长选择，并利用空间域加权平均法对所选波长对应的图像进行融合。在光谱信息降维方面，分别构建煤和矸石的比值光谱指数（RSI）、归一化光谱指数（NDSI）和光谱一阶微分光谱指数，根据煤和矸石的光谱指数归一化偏差值选择光谱指数；对比分析不同数据降维方法组合的煤矸识别准确率，得到：具有更高的识别准确率和更短的识别时间的组合是基于类间特征可分性的最佳指数法波段选择方法（[546nm，585nm，全色波段]组合波段）+光谱一阶微分指数构建方法（Area_{（622、661、702、PAN）}）。

（3）融合图像-光谱信息的煤矸识别方法研究。综合利用煤和矸石的多光谱图像信息和光谱信息，采用异构融合网络构建方法，构建一种能同时接受二维图像和一维光谱数据的多输入卷积神经网络煤矸识别模型。模型主要包括由残差网络和残差多尺度特征融合模块（ResMSFF）组成的图像特征提取模块、由残差网络和自注意力机制组成的光谱特征提取模块、由拼接融合模块和自注意力机制组成的特征融合模块，对模型进行训练测试，其分类精度达到99.17%。与只利用图像信息的煤矸识别准确率进行对比分析，得到：融合图像-光谱信息的煤矸识别模型具有更高的分类精度，并在低光照下仍有较高的识别准确率。

An important part of the coal production process is gangue sorting, and the industry has paid great attention to image-based gangue sorting technology due to its many advantages. The challenging conditions of the coal production process make it hard to acquire high-quality photos, and the physical similarities between coal and gangue further complicate identification. Richer data can be obtained by using multi-spectral imaging technology, which can simultaneously acquire the target object's picture and spectral information. In this paper, the technology is applied to coal and gangue sorting, and research on effects of illuminance and spectral band changes on coal and gangue image and spectral parameters, band combination selection method and identification method. The specific work is as follows:

(1) Analysis of the effect of illuminance and waveband on coal and gangue image and spectral features. Design the experimental program for acquiring multispectral images and spectral information data of coal and gangue in various illuminance and spectral bands. Create an experimental platform to create datasets of multispectral images and spectral values of coal and gangue in two mining regions under five different light intensities and nine spectral bands; analyzes image parameters (standard deviation, skewness, entropy, and contrast) and the spectral parameter, and compares the differences between coal and gangue parameters values. obtained: the multispectral images and spectral features of coal and gangue show different variations with the change of illuminance and spectral bands; the deviations in both image and spectral parameters are larger at 7000lux and 585nm than at other settings.

(2) Research of data dimensionality reduction methods based on the selection of spectral band combinations. For image dimensionality reduction, principal component analysis (PCA) and optimal index method (OIF) based on interclass feature distinguishability are used to select feature wavelengths for multispectral images, respectively, and the images corresponding to the selected wavelengths are fused using spatial-domain weighted averaging. For spectral information dimensionality reduction, the ratio spectral index (RSI), normalized spectral index (NDSI) and spectral first-order differential spectral index of coal and gangue were constructed respectively, and the spectral index was selected according to the normalized deviation value of spectral index of coal and gangue. Comparison the recognition accuracy of the coal and gangue of different combinations of data dimensionality reduction methods, obtained: the combination with higher recognition accuracy and shorter recognition time is the optimal exponential method band selection method based on the differentiability of interclass features ([546nm, 585nm, panchromatic bands] combination of the bands) + spectral first-order differential index construction method (Area _{(622, 661, 702, PAN)}).

(3) Research on the identification method of coal gangue by fusing image-spectrum information. By employing the heterogeneous fusion network construction method, a multi-input convolutional neural network coal and gangue recognition model that can simultaneously accept two-dimensional image and one-dimensional spectral data is built, utilizing multi-spectral image information and spectral information of coal and gangue. The model is comprised of three main modules: an image feature extraction module with residual network and residual multiscale feature fusion module (ResMSFF); a spectral feature extraction module with residual network and self-attention mechanism; and a feature fusion module with splicing and fusion module and self-attention mechanism. Based on training and testing, the model's classification accuracy is 99.17%. Comparative with the accuracy of coal and gangue identification using only image information, obtained: the coal and gangue identification model fusing image-spectral information has higher classification accuracy and still has high recognition accuracy under low light.

[1]谢和平，王金华，王国法，等．煤炭革命新理念与煤炭科技发展构想[J]．煤炭学报，2018，43(05)：1187-1197．

[2]于鹏．俄乌冲突对中国能源贸易的影响及对策研究[J]．价格月刊，2022，(09)：73-77．

[3]王双明，段中会，马丽等．西部煤炭绿色开发地质保障技术研究现状与发展趋势[J]．煤炭科学技术，2019，47(2)：1-6．

[4]姚逸，邓秋婷，李艺，卢海，邹荣灿．煤矸石的综合治理及其开发利用现状[J]．中国资源综合利用，2019，37(12)：83-85．

[5]张秀捧，郑剑平．国有重点煤炭企业“采选充一体化”技术调研及未来走向研判[J/OL]．煤炭加工与综合利用，2023(1)：1-11+15．

[6]钱鸣高．煤炭的科学开采[J]．煤炭学报，2010，35(04)：529-534．

[7]谢和平，任世华，谢亚辰，焦小淼．碳中和目标下煤炭行业发展机遇[J]．煤炭学报，2021，46(07)：2197-2211．

[8]谢和平，吴立新，郑德志．2025年中国能源消费及煤炭需求预测[J]．煤炭学报，2019，44(07)：1949-1960．

[9]刘虹．俄乌冲突下欧盟煤炭困局与思考[J]．煤炭经济研究，2022，42(03)：1．

[10]Jiayan L, Jinman W. Comprehensive utilization and environmental risks of coal gangue: A review[J]. Journal of Cleaner Production, 2019, 239: 117946.

[11]徐志强，吕子奇，王卫东，等．煤矸智能分选的机器视觉识别方法与优化[J]．煤炭学报，2020，45(06)：2207-2216．

[12]Xinquan W, Shuang W, Yongcun G, et al.Dielectric and geometric feature extraction and recognition method of coal and gangue based on VMD-SVM[J]. Powder Technology, 2021, 392241-250.

[13]曹亦俊，刘敏，邢耀文，李国胜，罗佳倩，桂夏辉．煤矿井下选煤技术现状和展望[J]．采矿与安全工程学报，2020，37(01)：192-201．

[14]夏云凯，刘建荣，王寅林等．“双碳”目标下促进干法选煤技术应用研究[J]．中国煤炭，2022，48(07)：22-26．

[15]Yang D, Li J, Zheng K, et al.Impact-Crush Separation Characteristics of Coal and Gangue[J]. International Journal of Coal Preparation and Utilization, 2016, 38(3): 127-134.

[16]H. K Z, D. K G, J. Z Y, et al. Morphological characterization of crushed coal gangue by micro X-ray computed tomography scanning[J]. International Journal of Coal Preparation and Utilization, 2020, 42(5): 1-20.

[17]Ningbo Z, Changyou L. Radiation characteristics of natural gamma-ray from coal and gangue for recognition in top coal caving.[J]. Scientific reports, 2018, 8(1): 190.

[18]郭永存，何磊，刘普壮，王希．煤矸双能X射线图像多维度分析识别方法[J]．煤炭学报，2021，46(01)：300-309．

[19]张锦旺，何庚，王家臣．不同混合度下液体介入难辨别煤矸红外图像识别准确率[J]．煤炭学报，2022，47(03)：1370-1381．

[20]李曼，杨茂林，刘长岳，等．基于图像的煤矸分选中图像照度调节方法[J]．煤炭学报，2021，46(S2)：1149-1158．

[21]Wenhao L, Mengran Z, Feng H, et al.A study of Multispectral Technology and Two-dimension Autoencoder for Coal and Gangue Recognition[J]. IEEE Access, 2020, 81-1.

[22]李廉洁，樊书祥，王学文，李瑞，文小，王璐瑶，李博．高光谱成像的煤与矸石分类[J]．光谱学与光谱分析，2022，42(04)：1250-1256．

[23]赵宝国．基于多光谱成像的人体指甲特征信息检测与识别研究[D]．中北大学，2019．

[24]杨文府，刘珺，汪雯雯，刘小松，郝晓阳．基于多光谱遥感数据的生物多样性监测与评估[J]．光谱学与光谱分析，2023，43(04)：1282-1290．

[25]张华锋，王武，白玉荣等．多光谱成像无损识别冻融猪肉中危害级碎骨[J]．光谱学与光谱分析，2021，41(09)：2892-2897．

[26]郜世姣，关海鸥，马晓丹，王彦宏．大豆冠层多光谱图像提取方法[J]．光谱学与光谱分析，2022，42(11)：3568-3574．

[27]Elmasry G, Kamruzzaman M, Sun D-W, et al. Principles and Applications of Hyperspectral Imaging in Quality Evaluation of Agro-Food Products: A Review[J]. Critical Reviews in Food Science and Nutrition, 2012, 52(11): 999-1023.

[28]杨玮，于滈，李浩，等．基于多光谱图像的土壤有机质含量检测系统与APP研究[J]．农业机械学报，2023，54(09)：270-278．

[29]Álvaro Agustín Chávez Durán, Mariano García, Miguel Olvera Vargas, et al. Forest Canopy Fuel Loads Mapping Using Unmanned Aerial Vehicle High-Resolution Red, Green, Blue and Multispectral Imagery[J]. Forests, 2024, 15(2): 225.

[30]Tan M, Zhang X, Luo W, et al. Deep Learning Based Spatial Distribution Estimation of Soil Pb Using Multi-Phase Multispectral Remote Sensing Images in a Mining Area[J]. Land, 2023, 12(9): 1789.

[31]Hu J, Feng H, Wang Q, et al. Pretrained Deep Learning Networks and Multispectral Imagery Enhance Maize LCC, FVC, and Maturity Estimation[J]. Remote Sensing, 2024, 16(5): 784.

[32]Kang Y, Wang Y, Fan Y, et al. Wheat Yield Estimation Based on Unmanned Aerial Vehicle Multispectral Images and Texture Feature Indices[J]. Agriculture, 2024, 14(2): 167.

[33]Janak D, Sneha C, Rahul G, et al. Multispectral Imaging for Vein Localization and Contrast Enhancement[J]. Annual International Conference of the IEEE Engineering in Medicine and Biology Society. IEEE Engineering in Medicine and Biology Society. Annual International Conference, 2023, 20231-4.

[34]O'Connor C, McCarthy S, Kiely L, et al. Novel multispectral imaging to predict disease progression in pediatric morphea[J]. Pediatric dermatology, 2024, 41(2): 229-233.

[35]KiSeung L. Multispectral Food Classification and Caloric Estimation Using Convolutional Neural Networks[J]. Foods (Basel, Switzerland), 2023, 12(17): 3212.

[36]Raju B V, Sazonov E. Detection of Oil-Containing Dressing on Salad Leaves Using Multispectral Imaging[J]. IEEE Access, 2020, 886196-86206.

[37]Jia Z, Chen G. Study on Digital Image Inpainting Method Based on Multispectral Image Decomposition Synthesis[J]. International Journal of Pattern Recognition and Artificial Intelligence, 2019, 33(1): 1954004.

[38]Feng H, Mengran Z, Pengcheng Y, et al.Multispectral Imaging: A New Solution for Identification of Coal and Gangue[J]. IEEE Access, 2019, 7169697-169704.

[39]来文豪，周孟然，胡锋，等．基于多光谱成像和改进YOLO v4的煤矸石检测[J]．光学学报，2020，40(24)：72-80．

[40]Pengcheng Y, Xuyue K, Heng Z, et al. Target Recognition of Coal and Gangue Based on Improved YOLOv5s and Spectral Technology.[J]. Sensors (Basel, Switzerland), 2023, 23(10): 4911.

[41]Wenhao L, Mengran Z, Feng H, et al.A study of Multispectral Technology and Two-dimension Autoencoder for Coal and Gangue Recognition[J]. IEEE Access, 2020, 81-1.

[42]张楷鑫．基于光谱成像技术的煤岩特征信息检测与识别方法研究[D]．西安科技大学，2021．

[43]胡锋．基于多光谱成像和深度学习的煤矸识别研究[D]．安徽理工大学，2020．

[44]曹现刚，郝朋英，王鹏等．多因素光照条件下高质量煤矸图像获取方法研究[J]．煤炭科学技术，2023，51(01)：455-463．

[45]Man L, Xianli H, Yong D, et al. Experimental study on the influence of external factors on image features of coal and gangue[J]. International Journal of Coal Preparation and Utilization, 2022, 42(9): 2770-2787.

[46]沈宁，窦东阳，杨程，等．基于机器视觉的煤矸石多工况识别研究[J]．煤炭工程，2019(01)：120-125．

[47]Qikai X, Yang Y, Yongbin L. Research on the Strong Generalization of Coal Gangue Recognition Technology Based on the Image and Convolutional Neural Network under Complex Conditions.[J]. ACS omega, 2023, 8(43): 40309-40320.

[48]李博，王学文，庞尚钟，等．煤与矸石图像特征分析及试验研究[J]．煤炭科学技术，2022，50(08)：236-246．

[49]苏宝金，陈波，贺靖峰，等．煤与矸石灰度直方图的差异研究[J]．洁净煤技术，2011，17(06)：96-98．

[50]吴开兴，宋剑．基于灰度共生矩阵的煤与矸石自动识别研究[J]．煤炭工程，2016，48(02)：98-101．

[51]余乐．一种煤和煤矸石图像识别的新方法[J]．现代计算机(专业版)，2017(17)：66-70．

[52]陈雪梅，张晞，徐莉莉，等．煤与矸石分形维数的差异研究[J]．煤炭科学技术，2017，45(07)：196-199．

[53]李曼，段雍，曹现刚，等．煤矸分选机器人图像识别方法和系统[J]．煤炭学报，2020，45(10)：3636-3644．

[54]Gang C, Jie C, Yifan W, et al.A Coal Gangue Identification Method Based on HOG Combined with LBP Features and Improved Support Vector Machine[J].Symmetry, 2023, 15(1): 202-202.

[55]Xinquan W, Shuang W, Yongcun G, et al. Dielectric and geometric feature extraction and recognition method of coal and gangue based on VMD-SVM[J]. Powder Technology, 2021, 392241-250.

[56]Tripathy P D, Reddy R G K. Novel Methods for Separation of Gangue from Limestone and Coal using Multispectral and Joint Color-Texture Features[J].Journal of The Institution of Engineers (India): Series D, 2017, 98(1): 109-117.

[57]霍平，曾翰林，霍柯言．基于图像处理的煤/矸密度识别系统的研究[J]．选煤技术，2015，(02)：69-73．

[58]程刚，陈杰，何磊．基于LBP特征与SVM的煤矸识别方法研究[J]．煤炭技术，2023，42(10)：12-15．

[59]Man L, Kaikai S. An image recognition approach for coal and gangue used in pick-up robot[C]//2018 IEEE International Conference on Real-time Computing and Robotics (RCAR). Ieee, 2018: 501-507.

[60]薛光辉，李秀莹，钱孝玲，等．基于随机森林的综放工作面煤矸图像识别[J]．工矿自动化，2020，46(05)：57-62．

[61]鲁恒润，王卫东，徐志强，等．基于机器视觉的煤矸特征提取与分类研究[J]．煤炭工程，2018，50(08)：137-140．

[62]Minghao Z, Wenhao L. Coal gangue recognition based on spectral imaging combined with XGBoost[J]. PloS one, 2023, 18(1): e0279955.

[63]Su L, Cao X, Ma H, et al. Research on coal gangue identification by using convolutional neural network[C]//2018 2nd IEEE Advanced Information Management, Communicates, Electronic and Automation Control Conference (IMCEC). IEEE, 2018: 810-814.

[64]曹现刚，薛祯也．基于迁移学习的GoogLenet煤矸石图像识别[J]．软件导刊，2019，18(12)：183-186．

[65]王闰泽，郎利影，席思星．用于智能煤矸分选机器人的改进型VGG网络煤矸识别模型[J]．煤炭技术，2022，41(01)：237-241．

[66]李莹．基于深度学习的煤矸石目标检测方法研究[D]．西安科技大学，2020．

[67]Gao R, Sun Z, Li W, et al. Automatic Coal and Gangue Segmentation Using U-Net Based Fully Convolutional Networks[J]. Energies, 2020, 13(4): 829.

[68]Junhao J, Yanfeng H, Huijun Z, et al. Recognition and sorting of coal and gangue based on image process and multilayer perceptron[J]. International Journal of Coal Preparation and Utilization, 2023, 43(1): 54-72.

[69]郜亚松，张步勤，郎利影．基于深度学习的煤矸石识别技术与实现[J]．煤炭科学技术，2021，49(12)：202-208．

[70]李娟莉，魏代良，李博，等．基于深度学习轻量化的改进SSD煤矸快速分选模型[J]．东北大学学报(自然科学版)，2023，44(10)：1474-1480．

[71]韦小龙，王方田，刘超，等．基于CSPNet-YOLOv7目标检测算法的煤矸图像识别模型[J/OL]．煤炭科学技术，1-13[2024-05-13]．

[72]Fang L, Cao Z, Li R, et al. Research on Image Recognition of Coal and Gangue Based on Improved ResNet[C]//2023 8th International Conference on Image, Vision and Computing (ICIVC). IEEE, 2023: 699-704.

[73]Man L, Xianli H, Yong D, et al. Experimental study on the influence of external factors on image features of coal and gangue[J]. International Journal of Coal Preparation and Utilization, 2022, 42(9): 2770-2787.

[74]段雍．基于图像的煤矸识别和定位方法研究与实现[D]．西安科技大学，2020．

[75]Kaiming H, Xiangyu Z, Shaoqing R, et al. Deep residual learning for image recognition[C]//Proceedings of the IEEE conference on computer vision and pattern recognition. 2016: 770-778.

[76]Yu F, Koltun V. Multi-Scale Context Aggregation by Dilated Convolutions[J]. CoRR, 2015, 1511.07122.

[77]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(09)：9-18．

[78]高玉才，付忠广，王诗云，等．基于Bi-LSTM和自注意力机制的旋转机械故障诊断方法研究[J]．中国工程机械学报，2022，20(03)：273-278．

[79]Cui X, Zheng K, Gao L, et al. Multiscale Spatial-Spectral Convolutional Network with Image-Based Framework for Hyperspectral Imagery Classification[J]. Remote Sensing, 2019, 11(19): 2220-2220.