基于图像的煤矸分选中，高质量的煤矸图像获取是实现自动化，智能化煤矸分选的首要环节。针对由于生产环境中光照不理想，引起图像亮度变化，图像质量降低，以及煤矸在输送带上背景难以提取，重叠样本较多，对定位和识别造成干扰等问题，提出了一种煤矸图像照度增强方法和输送带背景下的图像分割方法，并开发了图像增强和分割定位系统。本文主要开展了以下研究工作：

(1)建立不同照度的图像样本集，研究基于Retinex算法的煤矸图像照度调节方法。模拟实际工况环境照度的变化，分别在2500 lux、4000 lux、5500 lux、7000 lux 4种不同照度下，采集韩城矿区的瘦煤和页岩图像，并建立图像样本集。在对比分析高斯、双边、快速引导3种图像滤波算法的处理效果的基础上，确定基于快速引导滤波的Retinex算法用于煤矸图像照度调节，分析对比了不同光照因子对图像亮度的调节效果。

(2)对比分析图像照度调节对特征参数的影响规律，研究最佳光照因子的选取方法。对原始煤和矸石灰度特征进行分析，得到标准差和熵两个特征的区分度更好。分析不同光照因子增强前后煤和矸石标准差和熵单一灰度特征参数以及联合灰度特征参数的变化规律，基于通过特征有效范围来判别特征区分度大小的原理，提出了重叠面积比例法作为最佳光照因子的选取方法，并对选取效果进行了验证。

(3)研究煤矸样本与其输送带背景的图像分割方法，以及边缘重叠煤矸样本的分割方法。通过对比聚类、最大类间方差、最大熵、Gabor滤波4种背景分割算法的分割效果，得到基于Gabor滤波器的背景分割方法具有更好的分割效果。采用角点检测和形态学处理相结合的方法，对边缘重叠样本进行分割。采用质心法对分割后的样本进行定位。完成对相机的标定，得到现实坐标系与像素坐标系的转换关系。采用MATLAB为开发平台，完成本文图像增强和分割算法的实现。

(4)搭建实验平台，以标准差和熵为输入向量训练LSSVM分类器，随机选取工况条件下的煤矸样本，通过实验对论文提出的各方法进行验证。对比分析图像增强算法对煤矸识别率的影响，得到：2500 lux,4000 lux,5500 lux,7000 lux 4种固定照度下，增强后图像相对原始图像的识别率分别增加了7.5%,8.0%,8.5%,2.0%；由暗到亮变化照度条件下，不同灰度值的图像增强后识别率也均有提升；测试煤矸样本定位的精度，得到：X、Y轴最大误差分别为12.5 mm和7.5 mm，平均误差分别为5.2 mm和3.7 mm；测试背景分割的准确性，得到：背景像素的平均错分率为4.6%。

关 键 词：煤矸识别；照度；Retinex；图像增强；图像分割

研究类型：应用研究

In image-based coal and gangue separation, the acquisition of high-quality coal and gangue image is the first step to realize automatic and intelligent coal and gangue separation. Aim to slove the problem of the lighting in the production environment is not ideal,change the image brightness and further decrease image quality, and the coal and gangue are difficult to extract on the background of conveyor belt, some samples is overlapping, it also created some interference to locate and identify.Proposed an image illumination enhancement method and an image segmentation method under the background of conveyor belt , and developed an image enhancement and segmentation positioning system.This paper has carried out the following research work:

(1)This paper builds an image dataset consist of different illuminance,study the image illuminance adjustment method based on Retinex algorithm.Builds an image dataset consist of lean coal and shale from Hancheng City mining area, and simulates environmental illuminance of the actual working conditions by using four illuminances including 2500lux, 4000lux, 5500lux, 7000lux.On the basis of comparative analysis about the processing effects of three image filtering algorithms, Gaussian Filtering, Bilateral Filtering and Fast Guided Filtering, Retinex algorithm based on Fast Guided Filtering was selected as the illumination adjustment algorithm of coal and gangue image.The adjustment effect of different illumination factors on image brightness is analyzed and compared.

(2) Study the influence rule of illuminance adjustment on image characteristic parameters and the selection method of the optimal illuminance factor.Analyzed the gray characteristics of coal and gangue image, get standard-deviation and entropy two characteristics are better in discrimination.Analyzed using the different illumination factor to enhance coal and gangue images ,the standard deviation and entropy of the single gray characteristic parameters and joint grayscale change rules of characteristic parameters, and based on the principle of features effective range which could discriminant the separability degree of characteristic , put forward the overlapping area ratio method as the method of choosing the best illumination factor, and the selection effect is verified.

(3) Study the segmentation method of coal and gangue under the background of conveyor belt, and the sample segmentation method after the edge of coal and gangue samples overlapped.By comparing the segmentation effects of k-means, Otsu, maximum entropy and Gabor filter, it is concluded that the background segmentation method based on Gabor filter has better segmentation effects.At the same time, the method of corner searching and morphological processing was used to separate the overlapping coal and gangue samples, and the mass center method was used to locate the separated coal and gangue samples. Completed the camera calibration, and the conversion relationship between the real coordinate system and the pixel coordinate system is obtained.Using MATLAB as the development platform, the realization of image enhancement and segmentation algorithm in this paper is completed.

(4) Build an experimental sample platform,using LSSVM as the classifier trained and validated using entropy and standard-deviation as the input vectors, and test the identification and positioning of coal and gangue samples under actual working conditions to verify the image enhancement and image segmentation algorithm proposed. Analyzed the influence of image enhancement algorithm on coal and gangue identification rate,the results show that the image recognition rate enhancing under the four illuminances is increased by 7.5%, 8.0%, 8.5%, 2.0% respectively,under the condition of varying illuminance from dark to light, the recognition rate of images with different gray values also improved after enhancement.Tested the positioning accuracy of coal and gangue samples,the maximum error of positioning on X and Y coordinate pixel is 12.5mm and 7.5mm respectively, the mean error of X and Y coordinate pixels is 5.2mm and 3.7 mm respectively,Tested the accuracy of background segmentation,the average misclassification rate of background pixels was 4.6%.

Keywords：Coal and Gangue Recognition; Intensity of Illumination; Retinex; Image Enhancement; Image Segmentation

Thesis    ：Application Research

[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] 曹湘洪.实现我国煤化工、煤制油产业健康发展的若干思考 [J]. 化工进 展,2011,30(01):80-87.

[3] 中国国土资源报. 国土资源部有关负责人解读《关于加快建设绿色矿山的实施意见》 [J]. 国土资源, 2017(6):28-29.

[4] 杨旋.建设绿色矿山守住绿水青山——部规划司有关负责人解读《关于加快建设绿 色矿山的实施意见》[J]. 国土资源通讯, 2017(10).

[5] 缪协兴,钱鸣高.中国煤炭资源绿色开采研究现状与展望[J]. 采矿与安全工程学 报, 2009, 26(1):1-14.

[6] 钱鸣高.资源与环境协调(绿色)开采[J].煤炭科技, 2006(1):1-7.

[7] 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.

[8] 张振红.我国干法选煤技术发展现状与应用前景[J].选煤技术,2019(01):43-47+52.

[9] 李宛霖. 煤矸石多金属提取过程及固化机理研究[D].昆明理工大学,2019.

[10] 王东辉,刘文礼,张学仁,等.煤泥水压滤滤饼孔隙结构特征研究[J].矿业科学学 报,2017,2(02):183-190.

[11] 姜晓琳,解强.利用粉煤灰中溶出的硅、铝定向合成沸石[J].矿业科学学 报,2017,2(04):395-401.

[12] 王鹏涛.煤矸石综合利用的现状及存在的问题研究[J].科学技术创 新,2019(16):182-183.

[13] 米强,徐岩,刘斌,徐运杰.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.

[14] 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.

[15] 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.

[16] 孙继平,田子建.矿井图像监视系统与关键技术[J].煤炭科学技术,2014(01): 75-78.

[17] 孔锐,张冰.光照变化条件下人脸识别方法研究[J].系统仿真学报,2016,28(03):689-695.

[18] 宋世军,宋连玉,郝龙龙,乔彩风.光照变化对螺栓识别影响的实验研究[J].中国工程机 械学报,2012,10(03):348-352.

[19] 张坤,周东翔,彭科举,等.一种消除光照影响的舰船跑道区域增强方法[J].现代电子技 术,2014,37 (21):31-34.

[20] 屠珺,刘成良,李彦明,等.基于光照无关图的苹果图像识别方法[J].农业工程学 报,2010,26(S2):26-31.

[21] 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

[22] 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.

[23] Man Li, Xianli He, Yong Duan,et al.(2021) Experimental study on the influence of external factors on image features of coal and gangue, International Journal of Coal Preparation and Utilization, DOI: 10.1080/19392699.2021.1901692

[24] Mishro P K, Agrawal S, Panda R, et al. A novel brightness preserving joint histogram equalization technique for contrast enhancement of brain MR images[J]. Biocybernetics and Biomedical Engineering, 2021, 41(1).540-553,

[25] 张发军,彭文刚,虞成俊,何孔德.对比度受限下直方图均衡化处理的木板纹路图像增强 算法[J].科学技术与工程,2020, 20(21):8629-8636.

[26] Tiwari M,Bhupendra G.Brightness preserving contrast enhancement of medical images using adaptive gamma correction and homomorphic filtering[C]// 2016 IEEE Students' Conference on Electrical, Electronics and Computer Science. IEEE, 2016.

[27] 刘佳敏,何宁.基于改进同态滤波的低对比度图像增强[J].计算机应用与软 件,2020,37(03):220-224.

[28] 张志伟,叶建雄,程群,等.基于灰度变换的水下焊接图像对比度增强[J].热加工工 艺,2021,50(07):127-131.

[29] Priyadarshini M, Sasikala R, Meenakumari R.Novel approach for satellite image resolution and contrast enhancement using wavelet transform and Brightness Preserving Dynamic Histogram Equalization[C]// 2014 International Conference on Green Computing Communication and Electrical Engineering (ICGCCEE). IEEE, 2014.

[30]张英俊,雷耀花,潘理虎.基于暗原色先验的煤矿井下图像增强技术[J].工矿自动 化,2015,41(03):80-83

[31] J. Zhou, F.Zhou, "Single image dehazing motivated by Retinex theory," 2013 2nd International Symposium on Instrumentation and Measurement, Sensor Network and Automation (IMSNA), Toronto, ON, 2013, pp. 243-247, doi: 10.1109/IMSNA.2013. 6743260.

[32] J.Liu, J.Zhu, Y. Pei and Y. Zhang, "Adaptive depth map-based retinex for image defogging," 2016 International Conference on Audio, Language and Image Processing (ICALIP), Shanghai, 2016, pp. 318-322, doi: 10.1109/ ICALIP.2016.7846593.

[33] H. Hu and G. Ni, "Magetic resonance image enhancement based on multiscale Retinex algorithm," 2010 3rd International Conference on Biomedical Engineering and Informatics, Yantai, 2010, pp. 345-348, doi: 10.1109 /BMEI.2010.5639511.

[34] S.Setty, N.K. Srinath and M. C. Hanumantharaju, "Development of multiscale retinex algorithm for medical image enhancement based on multi-rate sampling," 2013 International Conference on Signal Processing , Image Processing & Pattern Recognition, Coimbatore, 2013, pp. 145-150, doi: 10.1109/ICSIPR.2013.6497976.

[35] W. Henan,Y.Guang, X. Zhonglin,et al. "Remote Sensing Image Enhancement Method Based on Multi-scale Retinex," 2011 International Conference of Information Technology, Computer Engineering and Management Sciences, Nanjing, Jiangsu, 2011, pp. 15-18, doi: 10.1109/ICM.2011.323.

[36] 蔡利梅,向秀华,李紫阳.自适应HSV空间Retinex煤矿监控图像增强算法[J].电视技 术,2017,41(Z1):11-15.

[37] 张立亚,郝博南,孟庆勇,等.基于HSV空间改进融合Retinex算法的井下图像增强方法 [J].煤炭学报,2020,45(S1):532-540.

[38] 智宁,毛善君,李梅.基于照度调整的矿井非均匀照度视频图像增强算法[J].煤炭学 报,2017,42(08):2190-2197.

[39] 王焱,关南楠,刘海涛.改进的多尺度Retinex井下图像增强算法[J].辽宁工程技术大学 学报(自然科学版),2016,35 (04):440-44

[40] 王梅,李玉鑑,全笑梅.图像分割的图论方法综述[J].计算机应用与软 件,2014,31(09):1-12+44.

[41] 杨蕴,李玉,赵泉华.基于局部空间信息的可变类模糊阈值光学遥感图像分割[J].自动化 学报,2022,48(02):582-593.

[42] 陈芳,张道强,廖洪恩,等.基于序列注意力和局部相位引导的骨超声图像分割网络 [J].自动化学报,2022,48:1-9.

[43] 邬博文,朱良宽,王璟瑀.基于差分进化鲸鱼优化算法的森林冠层图像分割[J].西北林学 院学报,2022,37(01):67-73+136.

[44] 马宇,张丽果,杜慧敏,等.卷积神经网络的交通标志语义分割[J].计算机科学与探 索,2021,15(06):1114-1121.

[45] 邓仕超,黄寅.二值图像膨胀腐蚀的快速算法[J].计算机工程与应 用,2017,53(05):207-211.

[46] Ma J, Le L.Hierarchical segmentation and identification of thoracic vertebra using learning-based edge detection and coarse-to-fine deformable model[J]. Computer Vision and Image Understanding, 2013, 117( 9):1072-1083.

[47] 魏光杏,周献中,李华.基于最大熵的进化区域生长分割算法[J].江汉大学学报(自然科 学版),2020,48(06):78-83.

[48] Bechar M, Settouti N, Daho M ,et al. Influence of normalization and color features on super-pixel classification: application to cytological image segmentation[J]. Australasian Physical & Engineering Sciences in Medicine, 2019, 42(2):427-441.

[49] 胡嵽,冯子亮.基于深度学习的轻量级道路图像语义分割算法[J].计算机应 用,2021,41(05):1326-1331.

[50] Khosravanian A, Rahmanimanesh M, Keshavarzi P, et al. Fast level set method for glioma brain tumor segmentation based on Superpixel fuzzy clustering and lattice Boltzmann method[J]. Computer Methods and Programs in Biomedicine, 2021, 198(17):105809.

[51] 尹建强,朱金波,曾秋予,杨晨光,张勇,史苘桧.X射线煤矸识别过程中图像分割精度研 究[J].选煤技术,2021(04):24-29.

[52] 王刚,陈雪畅,韩冬阳,等.基于改进Otsu的煤体CT图像阈值分割算法的研究[J].煤炭科 学技术,2021,49(01):264-271.

[53] Liu K, Zhang X, Chen Y. Extraction of Coal and Gangue Geometric Features with Multifractal Detrending Fluctuation Analysis. Applied Sciences. 2018; 8(3):463.

[54] 张泽琳,杨建国,苏晓兰.煤堆图像分割与特征提取[J].中南大学学报(自然科学 版),2014,45(06):1900-1907.

[55] Li M, Duan Y,He X, et al. Image positioning and identification method and system for coal and gangue sorting robot[J]. International Journal of Coal Preparation and Utilization, 2020(4):1-19.

[56] 曹现刚,刘思颖,王鹏,等.面向煤矸分拣机器人的煤矸识别定位系统研究[J].煤炭科学 技术,2022,50(01):237-246.

[57] 沈科,季亮,张袁浩,邹盛.基于改进YOLOv5s模型的煤矸目标检测[J].工矿自动 化,2021,47(11):107-111+118.

[58] Zou H, Jia R.Visual Positioning and Recognition of Gangues Based on Scratch Feature Detection[J]. Traitement du Signal, 2019, 36(2):147-153.

[59] LAND E, "Recent advances in retinex theory",Vision Research, vol. 26, no. 1, pp. 7-21, 1986.

[60] Jobson,D. J, Rahman, et al. A multiscale retinex for bridging the gap between color images and the human observation of scenes[J]. IEEE Transactions on Image Processing, 1997.

[61] Jiang B, Woodell G A, Jobson D J. Novel multi-scale retinex with color restoration on graphics processing unit[J]. Journal of Real-Time Image Processing, 2015.

[62] 李惠芬,蒋向前,李柱.高斯滤波稳健性能的研究与改进[J].仪器仪表学 报,2004(05):633-637.

[63] 杨九章,刘炜剑,程阳.基于对比度金字塔与双边滤波的非对称红外与可见光图像融合 [J].红外技术,2021,43(09):840-844.

[64] HE K, SUN J, TANG X. Guided image filtering[J]. IEEE Transactions on Pattern Analysis & Machine Intelligence,2011,35 (6) : 1397－1409.

[65] HE K, SUN J,Fast Guided Filter. arXiv Preprint arXiv: 1505. 00996, 2015

[66] A Horé, Ziou D.Image quality metrics: PSNR vs. SSIM[C]// 20th International Conference on Pattern Recognition, ICPR 2010, Istanbul, Turkey, 23-26 August 2010. IEEE Computer Society, 2010.

[67] Maeda K, Utsu M, Yamamoto N, et al. Echogenicity of fetal lung and liver quantified by the grey-level histogram width[J]. Ultrasound in Medicine & Biology, 1999, 25(2):201-208.

[68] GONZALEZ R,WOODS R,EDDINS S,et al. 数字图像处理的MATLAB实现[M]. 清华 大学出版社, 2013.447-448

[69] Mahsa, Mirzargar, Ross, et al. Curve Boxplot: Generalization of Boxplot for Ensembles of Curves[J]. IEEE Transactions on Visualization & Computer Graphics, 2014.

[70]孙凯凯. 基于图像的煤矸识别方法研究[D].西安科技大学,2018.

[71] S.Piramuthu, The Housdorff distance measure for feature selection in learning applications, in: Proceedings of the 32nd Hawaii International Conference on System Sciences, 1999, pp 1–6.

[72] K. Kira, L.A.Rendell, The feature selection problem: traditional methods and a new algorithm, in: Proceedings of the 9th National Conference on Artifificial Intelligence, 1992, pp 129–134.

[73] J.Liang, S.Yang, A.Winstanley, Invariant optimal feature selection: a distance discriminant and feature ranking based solution, The Journal of Pattern Recognition (2008) 1429–1439.

[74] P.E. Black (Ed.),Manhattan Distance, U.S. National Institute of Standards and Technology, 2006.

[75] Chandra B, Gupta M. An efficient statistical feature selection approach for classification of gene expression data[J]. Journal of Biomedical Informatics, 2011, 44( 4):529-535.

[76] 宋娟,黄勇,冯妍.基于相关匹配与面积重叠度的对流云跟踪方法[J].计算机工 程,2011,37(S1):336-337+340.

[77] 方园. 基于低秩模型和区域重叠法的胃部CT图像淋巴结识别[D].西安电子科技大 学,2014.

[78] 李兰,王朝立.基于局部梯度和面积重叠合并法的人脸检测[J].计算机仿 真,2014,31(05):279-283.

[79] Oh S.A new dataset evaluation method based on category overlap[J]. Computers in Biology & Medicine, 2011, 41(2):115-122.

[80] Wang J, Zhou S, Yi Y, et al. An improved feature selection based on effective range for classification.[J]. Scientific world journal, 2014, 2014:972125.

[81]Spackman K A.Signal Detection Theory: Valuable Tools for Evaluating Inductive Learning[J]. Morgan Kaufmann Publishers Inc. 1989.

[82] Grigoryev S G, Lobzin Y V, Skripchenko N V.The role and place of logistic regression and ROC analysis in solving medical diagnostic task[J]. Journal Infectology, 2016, 8(4):36-45.

[83] Omar L, Ivrissimtzis I.Using theoretical ROC curves for analysing machine learning binary classifiers[J]. Pattern recognition letters, 2019, 128(Dec.):447-451.

[84] Tappi S, Rocculi P, Ciampa A,et al.Computer vision system (CVS):a powerful non-destructive technique for the assessment of red mullet (Mullus barbatus) freshness[J]. European Food Research and Technology, 2017, 243(12):1-9.

[85] Hartigan J A, Wong M A . A K-Means Clustering Algorithm[J]. Applied Statistics, 1979, 28(1).

[86] 卜文斌,游福成,李泉,等.一种基于大津法改进的图像分割方法[J].北京印刷学院学 报,2015,23(04):76-78+82.

[87] 刘秀峰,庹先国,陆景,等.基于最大熵分割和区域生长的水岸线快速提取方法[J].制造 业自动化,2021,43(06):95-98+148.

[88] Gabor D.Theory of communication[J]. Iee Proc London, 1946, 93(73):58.

[89] Jain A K.Unsupervised texture segmentation using Gabor filters[J]. Pattern Recognition, 2002, 24.

[90] 武瑛.形态学图像处理的应用[J].计算机与现代化,2013(05):90-94.

[91] 蒋刚毅,郑义,郁梅,等.关于彩色图象处理的形态学算子研究[J].中国图象图形学 报,2000(04):27-33.

[92] Zafari S, Eerola T, Sampo J,et al.Segmentation of Partially Overlapping Nanoparticles Using Concave Points[C]// Springer International Publishing. Springer International Publishing, 2015.

[93] 徐 越,李盈慧,宋怀波,等.基于Snake模型与角点检测的双果重叠苹果目标分割方 法[J]. 农业工程学报,2015,31(1)：196-203.

[94]李娜,陈俊艳,刘颖,等.基于骨架角点检测的粘连车辆分割[J].西安邮电大学学 报,2015,20(06):14-18.DOI:10.13682/j.issn.2095-6533.2015.06.004.

[95] Harris C G, Stephens M J. A combined corner and edge detector[C] Alvey vision conference. 1988.

[96]Haralick, Robert M,and Linda G. Shapiro, Computer and Robot Vision, Volume I, Addison-Wesley, 1992, pp. 28-48.

[97]BresenhamJack.A linear algorithm for incremental digital display of circular arcs[J]. Communications of the Acm, 1977.

[98]汤荣志,段会川,孙海涛.SVM训练数据归一化研究[J].山东师范大学学报(自然科 版),2016,31(04):60-65.

[99] 刘鑫磊,梁耀,黎双文,等.复杂背景下重叠椭圆形叶片的分割算法[J].广东农业科 学,2019,46(05):149-157.