煤矸分选是煤炭生产过程中重要的工序环节，煤和矸石的准确识别和定位是煤矸分选机器人的首要任务。基于图像的煤矸识别方法劳动强度低、生产效率高、环境污染小，具有良好的发展前景。在煤矿实际生产中，复杂的外在因素易造成煤和矸石表面特征参数产生不同程度的变化，影响成像的清晰度，导致煤矸图像的有效差异特征难以被读取和识别，进而影响识别准确率。本文主要针对外在因素对图像特征的影响规律及多因素影响的图像识别方法进行研究，开展了以下工作：

（1）设计了照度、外在水分、皮带速度3种主要外在因素影响下煤和矸石图像采集实验方案，搭建了图像采集实验平台，建立了4种照度、7种外在水分、6种皮带速度下的煤矸图像样本集。对比分析了图像的4个灰度、4个纹理特征参数，得到了3种外在因素变化对煤矸图像特征参数的影响规律。对图像特征参数进行归一化处理，对比分析了煤与矸石各特征参数的偏差受照度、外在水分、皮带速度变化的波动情况。

（2）研究分析了粒子群优化算法(PSO)和XGBoost算法及其特点，提出了基于粒子群优化的XGBoost煤矸识别方法。通过对XGBoost模型影响较大参数值进行优化训练，得到各特征参数的重要性得分。分析不同特征参数组合对模型识别准确率的影响程度，得到以重要性得分前6个特征参数作为输入向量训练的模型准确率较高。与K-近邻法、最小二乘支持向量机识别模型进行对比实验，得到本文所提出的PSO-XGBoost算法具有更高的识别准确率，煤的识别率为93.5%，矸石的识别率为92.4%。

（3）研究分析了混合域注意力机制(CBAM)和YOLO算法及其优势，提出结合混合域注意力机制的YOLOv4煤矸识别方法。以多因素影响下的煤和矸石图像进行模型训练，通过测试得到添加注意力机制后的模型对煤、矸石、煤矸混合的识别准确率分别为98.2%、99.0%、97.3%，识别时间为32ms。分别与YOLOv4、LS-SVM、Fast R-CNN模型进行比较分析，得到YOLOv4+CBAM模型具有更优的识别准确率和识别时间。

（4）研究光照强度、外在水分和皮带速度3种外在因素对煤矸动态识别准确率的影响。采用正交实验法设计3因素4水平实验表，并生成正交实验方案。在实验室搭建动态测试实验平台，完成煤矸识别正交实验测试。对实验结果进行极差分析，分别得到3种因素对评判指标煤、矸石、煤矸混合识别准确率影响的权重顺序以及较优的因素水平组合方案，进一步分析各因素水平对评判指标的影响。

（5） 随机选取实际工况下的煤和矸石进行动态识别和定位验证。搭建实验平台，设计煤矸样本动态识别及定位程序。采用PSO-XGBoost模型得到煤样本识别准确率为93%，矸石样本的识别准确率为91%。采用YOLOv4+CBAM模型得到煤样本识别准确率为96%，矸石样本识别准确率为98%，煤矸混合样本识别准确率为95%；动态定位平均误差X、Y坐标分别为5.6mm、7.3mm；煤矸样本动态识别和定位的平均时间为55.9ms。

关 键 词：煤矸分选；灰度和纹理；外在因素；XGBoost；YOLOv4；正交实验

研究类型：应用研究

Coal and gangue sorting is a necessary process in coal production. Accurate recognition and location of coal and gangue is the primary task of the coal and gangue sorting robot. The image-based recognition method of coal and gangue has low labor intensity, high production efficiency, serious environmental pollution, and has a good development prospect. In the actual production, complex external factors cause different degrees of changes in the surface parameters of coal and gangue. It affects the sharpness of the image and make the effectively differentiated features hard to get and recognize, thus reducing the recognition accuracy. This paper mainly studies the influence law of external factors on image features and the image recognition method influenced by multiple factors, and carries out the following work.

(1)The experimental scheme of coal and gangue image acquisition under the influence of three main external factors, namely illumination, external moisture and belt speed, is designed. We build an experimental platform for image acquisition and construct an image dataset covering the scenarios of four illuminances, seven moistures, and six belt speeds. Analyzed image feature parameters of four grays and four textures and obtained the influence rules of three external factors on the feature parameters of the image. Through normalization, the fluctuation of deviation illumination, external moisture and belt velocity of coal and gangue are compared and analyzed.

(2)The particle swarm optimization algorithm (PSO) and XGBoost algorithm and their characteristics were analyzed, and an XGBoost method based on particle swarm optimization was proposed for coal and gangue recognition. Obtained the importance score of feature parameters by optimizing the parameter values of XGBoost. Analyzing the influence of different feature parameter combinations for model recognition accuracy, we get the top six parameters as input vector training had higher accuracy. Compared with the kNN and LS-SVM, the PSO-XGBoost has higher recognition accuracy, and the accuracy of coal and gangue are 93.5% and 92.4% respectively.

(3)The mixed domain attention mechanism (CBAM) and YOLO algorithm and their advantages were analyzed, and a YOLOv4 method combining the mixed domain attention mechanism was proposed to recognize coal and gangue. Training model using coal and gangue dataset under multiple factors influence. We obtained the recognition accuracy of coal, gangue, coal and gangue mixture was 98.2%, 99.0% and 97.3% respectively, and the recognition time was 32ms. Compared with YOLOv4, LS-SVM and Fast R-CNN models, the YOLOv4+CBAM model has better recognition accuracy and recognition time.

(4)The effects of light intensity, external moisture and belt velocity on dynamic identification accuracy of coal and gangue were studied. Designed the orthogonal experiment table of 3 factors and 4 levels, and generate the orthogonal experiment scheme. Build the dynamic test platform, and complete the orthogonal experimental of coal and gangue. Through the range analysis of the results, we obtained the weight order of the influence of three factors on the recognition accuracy of coal, gangue, coal and gangue mixture and the optimal combination the factor, futher analyzed the influence level of each factors on the evaluation index.

(5)Coal and gangue under actual working conditions are randomly selected for dynamic recognition and location verification.. Build the experimental platform and design the dynamic recognition and location program of coal and gangue. The recognition accuracy of coal and gangue is 93% and 91% respectively by the PSO-XGBoost. The dynamic recognition accuracy of coal, gangue and coal and gangue mixture are 96%, 98% and 95% respectively by the YOLOv4+CBAM. The average location error of the X and Y coordinate is 5.6mm and 7.3mm respectively. The average time of dynamic recognition and positioning of coal and gangue is 55.9ms.

Key words: Coal and gangue sorting; Grayscale and texture; External factors; XGBoost; YOLOv4; Orthogonal experimental   

Thesis    : Application Research

[1]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( 7) : 1949－1960．

[2]王国法. 加快煤矿智能化建设,推进煤炭行业高质量发展[J/OL]. 中国煤炭:1-9[2021-01-04].

[3]许佳林. 煤矿安全管理问题及措施分析[J].能源与节能,2020(12):157-158.

[4]孙玲,宫立昊. 2019年国内煤矿安全事故统计分析及对策研究[J]. 决策探索(中),2020(02):20-21.

[5]黄继广,马汉鹏,范春姣,等. 我国煤矿安全事故统计分析及预测[J]. 陕西煤炭,2020,39(03):34-39+6.

[6]谢和平，王金华，鞠杨，等．煤炭革命的战略与方向[J]．北京: 科学出版社，2018: 1-31.

[7]杨方亮．煤炭资源综合利用发电现状分析与前景探讨[J]．中国煤炭，2020，46(10):67-74.

[8]李小炯. 我国煤炭资源清洁高效利用现状及对策建议[J]. 煤炭经济研究,2019,39(01):71-75.

[9]Zhu Xueshuai, Liu Junli, Yang Xizu,et al. Rheological properties of dense medium suspension for coal preparation[J]. Energy Sources Part A Recovery Utilization and Environmental Effects,2020.

[10]曹亦俊,刘敏,邢耀文,等. 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.

[11]Paul Subharanjan, Bhattacharya Sumantra. Size by Size Separation Characteristics of a Coal Cleaning Jig[J]. Transactions of the Indian Institute of Metals,2018,71(6).

[12]Zhao Yiding, He Xiaoming. Recognition of Coal and Gangue Based on X-Ray[J]. Applied Mechanics and Materials, 2013, 2212 (275-277).

[13]商德勇,章林,牛艳奇,等. 煤矸分拣机器人设计与关键技术分析[J/OL]. 煤炭科学技术,1-7.

[14]王鹏,曹现刚,马宏伟,等. 基于余弦定理-PID的煤矸石分拣机器人动态目标稳准抓取算法[J]. 煤炭学报,2020,45(12):4240-4247.

[15]余乐,郑力新,杜永兆,等. 采用部分灰度压缩扩阶共生矩阵的煤和煤矸石图像识别[J]. 华侨大学学报(自然科学版),2018,39(06):906-912.

[16]曹现刚,费佳浩,王鹏,等. 基于多机械臂协同的煤矸分拣方法研究[J]. 煤炭科学技术,2019,47(04):7-12.

[17]曾翰林. 基于图像处理的煤矸识别技术研究[D]. 华北理工大学,2015.

[18]刘富强,钱建生,王新红,等. 基于图像处理与识别技术的煤矿矸石自动分选[J]. 煤炭学报,2000(05):534-537.

[19]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.

[20]Xu Jie, Wang Fengen. Study of Automatic Separation System of Coal and Gangue by IR Image Recognition Technology[J]. Advances in Automation and Robotics,2011, 2:87-92.

[21]于国防,邹士威,秦聪. 图像灰度信息在煤矸石自动分选中的应用研究[J]. 工矿自动化,2012,38(02):36-39.

[22]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.

[23]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.

[24]Li Man, Sun Kaikai. An image recognition approach for coal and gangue used in pick-up robot[J]. 2018 IEEE International Conference on Real-time Computing and Robotics(RCAR), 2018.

[25]谭春超,杨洁明. 煤与矸石图像灰度信息和纹理特征的提取研究[J]. 工矿自动化,2017,43(04):27-31.

[26]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(7):196–199.

[27]米强,徐岩,刘斌,等. 煤与矸石图像纹理特征提取方法[J]. 工矿自动化,2017,43(05):26-30.

[28]伍云霞,田一民. 基于字典学习的煤岩图像特征提取与识别方法[J]. 煤炭学报,2016, 41(12):3190-3196.

[29]孙继平,佘杰. 基于小波的煤岩图像特征抽取与识别[J]. 煤炭学报,2013,38(10):1900-1904.

[30]孙继平,杨坤. 一种煤岩图像特征提取与识别方法[J]. 工矿自动化,2017,43(05):1-5.

[31]Liu Kai, Zhang Xi, Chen Yangquan. Extraction of Coal and Gangue Geometric Features with Multifractal Detrending Fluctuation Analysis[J]. Applied Sciences,2018, 8(03).

[32]Tripathy Debi Prasad , Reddy K. Guru Raghavendra . Novel methods for separation of gan-gue from limestone and coal using multispectral and joint color-texture features[J]. Journal of The Institution of Engineers (India): Series D, 2017, 98(01): 109-117.

[33]Eshaq Refat Mohammed Abdullah, Hu Eryi, Li Menggang,et al. Separation Between Coal and Gangue Based on Infrared Radiation and Visual Extraction of the YCbCr Color Space[J]. IEEE Access,2020,8.

[34]曾翰林. 基于图像处理的煤矸识别技术研究[D]. 华北理工大学, 2015.

[35]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(2): 69-73.

[36]何敏,王培培,蒋慧慧. 基于SVM和纹理的煤和煤矸石自动识别[J]. 计算机工程与设计,2012,33(03):1117-1121.

[37]Dou Dongyang, Wu Wenze, Yang Jianguo, et al. Classification of coal and gangue under multiple surface conditions via machine vision and relief-SVM[J]. Powder Technology,2019(356):1024- 1028.

[38]武国平,梁兴国,胡金良,等. 图像处理和SVM应用于煤矸石分选的实验研究[J]. 信息技术,2019(1):97-102,107. 

[39]Wang Weidng, Lv Ziqi, Lu Hengrun. (2018). Research on methods to differentiate coal and gangue using image processing and a support vector machine. International Journal of Coal Preparation and Utilization, 1-14.

[40]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.

[41]郭永存,于中山,卢熠昌. 基于PSO优化NP-FSVM的煤矸光电智能分选技术研究[J].煤炭科学技术,2019,47(04):13-19.

[42]Wang Bingjun, Huang Haoxiang, Dou Dongyang,et al. (2021). Detection of coal content in gangue via image analysis and particle swarm optimization-support vector machine. International Journal of Coal Preparation and Utilization, 1-10. 

[43]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.

[44]Li Man, Duan Yong, HE Xianli,et al. Image positioning and identification method and system for coal and gangue sorting robot[J]. International Journal of coal preparation and Utilization,2020.

[45]洪惠超. 基于机器视觉的煤矸石分选算法的研究[D]. 华侨大学, 2018.

[46]谭春超. 基于图像处理技术的煤矸识别与分选技术研究[D].太原理工大学,2017.

[47]曹现刚,李莹,王鹏,等.煤矸石识别方法研究现状与展望[J]. 工矿自动化,2020,46(01):38-43.

[48]Su Lingling , Cao Xiangang, Ma Hongwei,et al. Research on Coal Gangue Identification by Using Convolutional Neural Network[J]. 2018 2nd IEEE Advanced Information Management, Communicates, Electronic and Automation Control Conference (IMCEC), Xi'an, 2018, pp. 810-814.

[49]曹现刚,费佳浩,王 鹏,等. 基于多机械臂协同的煤矸分拣方法研究[J]. 煤炭科学技术,2019,47(4):7-12.

[50]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.

[51]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.

[52]单鹏飞,孙浩强,来兴平,等. 基于改进Faster R-CNN的综放煤矸混合放出状态识别方法[J/OL]. 煤炭学报:1-13[2022-03-03].

[53]Li Dongjun, Zhang Zhenxin, Xu Zhihua, et al. An Image-Based Hierarchical Deep Learning Framework for Coal and Gangue Detection[J]. IEEE Access, 2019,7, 184686-184699.

[54]宋卫虎. 基于深度学习的煤矸识别研究与实现[D]. 河北工程大学,2021.

[55]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.

[56]饶中钰,吴景涛,李明. 煤矸石图像分类方法[J]. 工矿自动化,2020,46(03):69-73.

[57]郭永存,王希,何磊,刘普壮. 基于TW-RN优化CNN的煤矸识别方法研究[J]. 煤炭科学技术,2022,50(01):228-236.

[58]高新宇. 基于机器视觉的煤矸智能分选系统设计[D].太原理工大学,2021.

[59]杜京义,史志芒,郝乐,等. 轻量化煤矸目标检测方法研究[J]. 工矿自动化,2021,47(11):119-125.

[60]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(03):50-55.

[61]Li Deyong, Wang Guofa, Zhang Yong,et al. Coal gangue detection and recognition algorithm based on deformable convolution YOLOv3[J]. IET Image Processing,2021,16(1).

[62]Yang Lei, Luo Jianchen, Song Xiaowei,et al. Robust Vehicle Speed Measurement Based on Feature Information Fusion for Vehicle Multi-Characteristic Detection[J]. Entropy,2021,23(7).

[63]谢斌红,袁帅,龚大立. 基于RDB-YOLOv4的煤矿井下有遮挡行人检测[J/OL]. 计算机工程与应用:1-10[2021-06-10].

[64]侯涛,蒋瑜. 改进YOLOv4在遥感飞机目标检测中的应用研究[J]. 计算机工程与应用,2021,57(12):224-230.

[65]Fu Huixuan, Song Guoqing, Wang Yuchao. Improved YOLOv4 Marine Target Detection Combined with CBAM[J]. Symmetry,2021,13(4),623.

[66]Bai Tangbo, Gao Jialin, Yang Jianwei,et al. A Study on Railway Surface Defects Detection Based on Machine Vision[J]. Entropy,2021,23(11).

[67]李彬,汪诚,吴静,等. 改进YOLOv4算法的航空发动机部件表面缺陷检测[J/OL]. 激光与光电子学进展:1-17[2021-06-10].

[68]刘新慧,吴燕萍. 煤炭水分问题对煤质管理的影响[J]. 内蒙古煤炭经济,2015(02):131-132.

[69]王家臣,李良晖,杨胜利. 不同照度下煤矸图像灰度及纹理特征提取的实验研究[J]. 煤炭学报,2018,43(11):3051-3061.

[70]苏晓兰. 选煤厂煤样图像采集系统设计与实现[D]. 中国矿业大学,2014.

[71]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研究[J/OL]. 煤炭学报:1-8[2021-04-16].

[72]曹现刚,郝朋英,王鹏,等. 多因素光照条件下高质量图像获取方法研究[J/OL]. 煤炭科学技术:1-11[2022-03-03].

[73]王新华,夏云凯. 原煤水分对干选效果的影响[J]. 煤炭加工与综合利用,2019(11):41-45,48.

[74]Li Man, He Xianli, Duan Yong,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, 2021.

[75]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.

[76]赵巧蓉. 差异光照对煤矸特征参数及识别精度影响研究[D]. 河北工程大学,2021.

[77]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研究[J/OL]. 煤炭学报:1-13[2022-03-03].

[78]张锦旺,王家臣,何庚,等. 液体介入提升煤矸识别效率的试验研究[J/OL]. 煤炭学报:1-14[2022-03-03].

[79]赵玺. 基于集成学习的混合气体分类和浓度预测算法研究[D]. 哈尔滨工业大学,2019.

[80]宋维. 基于空间和通道双重注意力机制的变电站施工安全带分割方法研究[D]. 华北电力大学,2020.

[81]罗勇琛. 融合空间和通道注意力机制的害虫图像识别方法研究[D]. 吉林农业大学,2021.

[82]肖振久,杨玥莹,孔祥旭. 基于改进YOLOv4的遥感图像目标检测方法[J/OL]. 激光与光电子学进展:1-14[2022-03-08].

[83]任丰仪,裴信彪,乔正,白越,等. 融合CBAM的YOLOv4轻量化检测方法[J/OL]. 小型微型计算机系统:1-8[2022-03-08].

[84]高金彪,徐德龙,王秀明,等. 基于正交设计的超声波稠油降黏实验[J]. 石油化工,2021,50(04):313-318.

[85]汪永明,李偎,谈莉斌,胡继涛,董书豪.基于正交实验的弯链板U弯冲压成形数值仿真[J].锻压技术,2021,46(12):46-53.