基于机器视觉的块矸石含煤率在线检测方法研究.pdf

硕士学位论文 基于机器视觉的块矸石含煤率 在线检测方法研究 Research on Online Detection of Coal Content of Massive Gangue Based on Machine Vision 作 者周德炀 导 师窦东阳副教授 中国矿业大学 二〇二一年五月 万方数据 学位论文使用授权声明学位论文使用授权声明 本人完全了解中国矿业大学有关保留、使用学位论文的规定，同意本人所撰 写的学位论文的使用授权按照学校的管理规定处理 作为申请学位的条件之一， 学位论文著作权拥有者须授权所在学校拥有学位 论文的部分使用权，即①学校档案馆和图书馆有权保留学位论文的纸质版和电 子版，可以使用影印、缩印或扫描等复制手段保存和汇编学位论文；②为教学和 科研目的，学校档案馆和图书馆可以将公开的学位论文作为资料在档案馆、图书 馆等场所或在校园网上供校内师生阅读、浏览。另外，根据有关法规，同意中国 国家图书馆保存研究生学位论文。 （保密的学位论文在解密后适用本授权书）。 作者签名 导师签名 年 月 日 年 月 日 万方数据 中图分类号 TD94 学校代码 10290 UDC 622 密 级 公开 中国矿业大学 硕士学位论文 基于机器视觉的块矸石含煤率 在线检测方法研究 Research on Online Detection of Coal Content of Massive Gangue Based on Machine Vision 作 者 周德炀 导 师 窦东阳 申请学位 工学硕士学位 培养单位 化工学院 学科专业 化工过程机械 研究方向 过程控制 答辩委员会主席 杨建国 评 阅 人 二○二一年 五月 万方数据 致谢致谢 七年岁月匆匆而过，又一次在矿大这个百年学府中毕业，却也是和矿大说再 见的时候。忆往昔，纵然此时此刻心中有万千个不舍，但时光不复，我能做的只 是心存感谢，勇往直前。在这论文定稿之际，向那些一直以来关心帮助我的老师 、同学以及默默支持我的家人至以最真诚的感谢。 首先由衷的感谢我的恩师窦东阳副教授。窦老师和蔼可亲、性情洒脱，与之 相处如浴春风。在矿大的求学历程中，窦老师在学习与生活中给予我悉心教导与 关怀，窦老师严谨治学的态度让我在研究生期间脚踏实地，做好科研的每一步。 论文的完成离不开窦老师帮助，在选题、实验、撰写论文环节，每个过程都离不 开窦老师的谆谆教导。在我的学术生涯中，窦老师既是启蒙者，也是助力者，从 本科毕业到如今的硕士毕业，我在大学历程中最重要的两个阶段，窦老师赋予我 双翼，教予我飞翔。而今我将飞向远方，在此衷心的感谢窦老师培育之恩，祝老 师身体康健，事业如意，桃李满园。 感谢课题组内杨建国老师、王羽玲老师、沈利民老师和王艳飞老师等在学习 与生活中指导和帮助。 感谢团队中沈晓峰和王立鹏在学习上的合作与生活中的帮 助。感谢薛妍、邱照玉、刘钢洋、王丙钧、黄浩翔和张纯龙等在毕业实验中提供 的帮助和生活中的鼓励。另外，感谢课题组郭志敏、姚宗序、周郑琪和李埴昂在 学习与生活中的帮助。再次，真诚的感谢课题组内的所有成员。 感谢室友王聪、高翰林、王景涛、杨俊秋、张伟胜和白沛瑶，让我在孤身求 学时有一个温暖的家。感谢红柳洗煤厂张立忠厂长、景治师兄等人在现场给予的 大力支持。 感谢我的父母，我的姐姐，感谢你们这么多年来的默默付出与支持，让我心 无旁骛的投身学业，取得如今的成绩，是你们成就了现在的我，你们是我坚强的 后盾。 最后，感谢百忙之中审阅论文的专家老师，感谢各位老师为论文提出批评指 正。 万方数据 I 摘摘 要要 矸石含煤率是煤炭分选领域中一项重要的技术考核指标， 反映了分选设备的 性能和操作人员的水平。目前该指标还不能实现在线检测，人为的检测方法不仅 消耗大量的人力物力，而且检测耗时过长不利于现场调控，这严重制约了选煤智 能化的发展，通过新方法实现矸石含煤率的实时检测迫在眉睫。机器视觉技术的 飞速发展，使之被广泛运用于煤炭洗选领域。基于现有的研究，针对矸石含煤率 暂未实现在线检测的现状，探索出一种基于机器视觉的矸石含煤率在线检测方 法。 从矸石含煤率的计算理论出发，通过图像处理技术、煤矸识别模型、煤矸体 积预测模型等研究探索出一种基于机器视觉的矸石含煤率在线检测方法。 在图像 处理技术使用过程中，首先选择基于高斯函数和 Hessian 矩阵的多尺度边缘检测 算法进行图像分割；又考虑到堆叠煤矸图像中因为遮挡造成分割区域的缺失，提 出了通过凸包算法补全缺失区域，来降低提取尺寸特征和形状特征时的误差。 对于煤矸识别模型，根据现有的研究成果选择出 12 种可被用于煤与矸石识 别的图像特征，将提取的特征组合为训练样本集，建立决策树识别模型，实验结 果表明决策树算法可以对煤与矸石进行有效识别， 并以此为基础通过集成学习建 立了基于 AdaBoost-决策树算法的煤矸识别模型，进一步提高识别的精度，识别 模型的 10 折交叉验证分类错误率最低为 3.29。 在煤与矸石的体积预测中，基于形状与体积之间的联系，提出根据形状分类 的体积回归预测模型。 首先依据煤与矸石的形状特征， 通过 K-Means聚类初步 离散各形状特征的特征值，再根据离散后的形状特征通过 K-Modes 聚类实现基 于形状特征对煤或矸石的分类。在现有对块煤的体积预测回归模型的研究基础 上，提出将二维尺寸特征转化为三维体积特征作为体积预测回归模型的自变量， 之后列出所有可以作为体积预测回归模型的自变量， 通过全子集回归找到最优回 归模型的自变量，结合回归 t 检验获得最终 5 个可用于体积预测回归模型的自变 量，并通过基于 bisquare 权重法的稳健回归降低异常样本的影响。根据煤或矸石 的形状分类结果，对不同形状类别分别建立体积回归模型，模型的平均相对误差 在 10左右，说明模型可以有效预测煤与矸石的体积。 在对图像处理技术、煤矸识别模型和煤矸体积预测模型研究的基础上，提出 基于机器视觉的矸石含煤率在线检测方法， 并通过实验室模拟对该方法进行效果 检验。在实验室模拟过程中，研究了煤与矸石样本的密度分布，在其不服从正态 分布的基础上使用 Bootstrap 法求解出密度均值及其置信区间，计算出使用密度 均值计算煤与矸石质量的误差分别在 2和 4左右。针对实验室模拟过程中得 到的矸石含煤率预测值远小于实际值的问题，梳理预测流程后发现，是分割后的 万方数据 II 区域普遍比实际区域偏大导致的，为降低分割造成的误差，提出加入补偿的矸石 含煤率计算公式，最终的实验室模拟的矸石含煤率预测的平均相对误差为 14.81，方法首次实现了矸石含煤率在线检测，预测精度尚令人满意。 该论文有图 40 幅，表 21 个，参考文献 91 篇。 关键词关键词机器视觉；矸石含煤率；煤矸识别；煤矸形状分类；煤矸体积预测 万方数据 III Abstract The coal content of gangue is an important technical assessment index in the field of coal sorting, which reflects the perance of the sorting equipment and the level of operators. At present, this indicator cannot be detected online. For man-made detection s, a lot of manpower and material resources are consumed, and detection takes too long to be unfavorable for on-site control, which severely restricts the development of intelligent coal preparation. Therefore, it is extremely urgent to realize real-time detection of the coal content of gangue through a new . The rapid development of machine vision technology has made it widely used in the field of coal washing. Based on the existing research, given the current situation that the coal content of gangue has not yet been detected online, a machine vision-based online detection for the coal content of gangue is explored. Starting from the theoretical calculation of the rate of coal containing waste rock, through image processing technology, coal gangue recognition model, coal gangue volume prediction model, etc., an online detection of gangue coal content based on machine vision is explored. In the process of using image processing technology, first, a multi-scale edge detection algorithm based on Gaussian function and the Hessian matrix is used for image segmentation; also considering the lack of segmentation area caused by occlusion in stacked Coal and gangue images, it is complemented by convex hull algorithm Missing regions are proposed to reduce the error in extracting dimensional features and shape features. For the Coal and gangue recognition model, 12 image features that can be used for coal and gangue recognition are selected based on existing research results. The extracted features are combined into a training sample set to build a decision tree recognition model. The experimental results show that coal and gangue can be effectively identified by the decision tree algorithm. Based on this, through ensemble learning, a Coal and gangue recognition model based on the AdaBoost-decision tree algorithm is established to further improve the accuracy of recognition. The 10-fold cross-validation classification error rate of the recognition model is at least 3.29. In the volume prediction of coal and gangue, based on the relationship between shape and volume, a volume regression prediction model based on shape classification is proposed. First, according to the shape characteristics of coal and gangue, through K-Means clustering, the feature value of each shape feature is initially discretized, and then according to the discrete shape feature, the classification 万方数据 IV of coal or gangue based on shape feature is realized through K-Modes clustering. Based on the existing research on the volume prediction regression model of lump coal, the transation of two-dimensional size features into three-dimensional volume features as the independent variable of the volume prediction regression model is proposed. After that, all the independent variables that can be used as the volume prediction regression model are listed. The independent variables of the optimal regression model are found through the full subset regression and combined with the regression t-test, 5 independent variables that can be used for the volume prediction regression model are obtained, and finally, Robust regression based on bisquare weighting is used to reduce the influence of abnormal samples. According to the shape classification results of coal or gangue, a volume regression model is established for different shape categories. The average relative error of the model is about 10, indicating that the volume model of coal and gangue can be predicted effectively. Based on the research of image processing technology, Coal and gangue recognition model, and Coal and gangue volume prediction model, an online detection of gangue coal content based on machine vision is proposed, and the effect of this is tested by laboratory simulation. In the laboratory simulation process, the density distribution of coal and gangue samples was studied. On the basis that it did not obey the normal distribution, the Bootstrap was used to solve the density mean and its confidence interval, and the density mean was used to calculate the coal and gangue. The quality error is about 2 and 4 respectively. Given the problem that the predicted value of the coal content of gangue obtained in the laboratory simulation process is much smaller than the actual value, after combing the prediction process, it is found that the divided area is generally larger than the actual area. To reduce the error caused by the division, compensation is added. The calculation ula for the coal content of the gangue in the final laboratory simulation is proposed. The average relative error of the prediction of the coal content of the gangue in the final laboratory simulation is 14.81, which is the first to realize the online detection of the gangue coal content, and the prediction accuracy is still satisfactory. Keywords Machine vision; Coal content of gangue; Identification of coal and gangue; Shape classification of coal and gangue; Volume prediction of coal and gangue 万方数据 V 目目 录录 摘摘 要要............................................................................................................................ I 目目 录录........................................................................................................................... V 图清单图清单......................................................................................................................... IX 表清单表清单........................................................................................................................ XII 变量注释表变量注释表 ..............................................................................................................XIV 1 绪论绪论............................................................................................................................ 1 1.1 研究背景及意义 ..................................................................................................... 1 1.2 研究内容 ................................................................................................................. 1 2 文献综述文献综述 ................................................................................................................... 4 2.1 机器视觉对矿物分选的应用研究 ......................................................................... 4 2.2 机器视觉在选煤领域的应用 ................................................................................. 8 2.3 机器视觉在煤矸识别上的研究现状 ..................................................................... 9 2.4 本章小结 ............................................................................................................... 13 3 煤和矸石的图像分割煤和矸石的图像分割 ............................................................................................. 14 3.1 图像分割 ............................................................................................................... 14 3.2 堆叠煤矸图片的区域补全 ................................................................................... 15 3.3 本章小结 ............................................................................................................... 18 4 煤与矸石的识别煤与矸石的识别 ..................................................................................................... 18 4.1 煤矸识别的特征提取 ........................................................................................... 19 4.2 集成学习的煤矸识别分类 ................................................................................... 24 4.3 煤矸识别模型的建立 ........................................................................................... 28 4.4 本章小结 ............................................................................................................... 34 5 煤与矸石的体积煤与矸石的体积预测预测 ............................................................................................. 35 5.1 煤矸体积预测特征的提取 ................................................................................... 35 5.2 煤与矸石的形状聚类分析 ................................................................................... 39 5.3 煤与矸石体积预测模型 ....................................................................................... 53 5.4 本章小结 ............................................................................................................... 63 6 矸石含煤率的在线检测矸石含煤率的在线检测 ......................................................................................... 64 6.1 密度的选取 ........................................................................................................... 64 6.2 矸石含煤率的在线检测 ....................................................................................... 66 万方数据 VI 6.3 工业现场应用 ....................................................................................................... 70 6.4 本章小结 ............................................................................................................... 72 7 总结与展望总结与展望 ............................................................................................................. 73 7.1 总结 ....................................................................................................................... 73 7.2 展望 ....................................................................................................................... 74 参考文献参考文献 ..................................................................................................................... 75 作者简历作者简历 ..................................................................................................................... 81 学位论文原创性声明学位论文原创性声明 ................................................................................................. 82 学位论文数据集学位论文数据集 ......................................................................................................... 83 万方数据 VII Contents Abstract ......................................................................................................................... I Contents ...................................................................................................................... VI List of Figures ............................................................................................................. IX List of Tables ............................................................................................................. XII List of Variables ....................................................................................................... XIV 1 Introduction ............................................................................................................... 1 1.1 Research Background and Significance ................................................................... 1 1.2 Research Content ..................................................................................................... 1 1.3 Summary .................................................................................................................. 3 2 Literature Review ..................................................................................................... 4 2.1 Research on the Application of Machine Vision to Mineral Separation .................. 4 2.2 Application of Machine Vision in the Field of Coal Preparation ............................. 8 2.3 Research Status of Machine Vision in Coal and Gangue Recognition .................... 9 2.4 Summary ................................................................................................................ 13 3 Image Segmentation of Coal a