煤炭杂物的高效检测是智能化选煤进程中的首要任务，也是提高分选率的关键环节。目前煤炭企业常用多结构化分离设备实现多类别煤炭杂物分选，该方式存在分选效率低、设备故障频发等多种安全隐患。因此，本文基于团队研发的煤杂分拣机器人及团队前期煤矸检测方法研究，实现对包括矸石在内的多种煤炭杂物目标检测，旨在解决复杂环境中煤炭杂物检测难题，为煤杂分拣机器人提供杂物检测信息。主要研究内容如下：

根据煤杂分拣机器人作业环境，分析视觉检测模块中相机、光源选型，搭建煤炭杂物视觉检测平台。针对煤炭杂物样本量缺乏的问题，采用网路爬虫、选煤厂采集及实验室模拟制作方式获取煤炭杂物样本，经过图像预处理构建复杂条件下煤炭杂物数据集。

为实现小样本煤炭杂物视觉检测，基于迁移学习方法对SSD目标检测模型进行迁移训练。根据煤炭杂物分拣机器人分拣需求构建基于SSD架构的煤炭杂物检测模型，与Faster-RCNN检测模型进行性能对比实验，结合煤炭杂物检测精度及效率平衡性需求，验证本文检测模型的可行性和可靠性。

为实现SSD煤炭杂物检测模型更好平衡检测精度与检测速度，同时解决煤炭杂物目标尺度多样性等问题，本文改进SSD煤炭杂物检测模型，构建轻量化煤炭杂物检测模型。首先分别用MobileNetV1和MobileNetV2特征提取网络替换原SSD中VGG16网络结构并进行对比试验；随后使用非极大抑制优化方法Soft-NMS改善煤炭杂物杂物检测框重叠、漏检问题；最后，利用MobileNetV2特有的反向残差结构进行模型剪枝优化，在保证模型精度损失较小的情况下进一步提升模型检测效率。

设计一煤炭杂物视觉检测系统，实现对带式输送机上的煤和杂物目标进行在线检测，该系统具有图像采集、预处理、杂物检测及可视化四部分功能。随后，使用550个煤炭杂物目标进行静态环境测试和动态实验测试，实验结果表明视觉检测系统满足机器人分拣需求，具有一定实用性

Efficient detection of coal debris is the primary task in the process of intelligent coal preparation, and it is also a key link to improve the sorting rate. At present, coal enterprises often use multi-structured separation equipment to achieve multi-category coal debris sorting. This method has many safety hazards such as low sorting efficiency and frequent equipment failures. Therefore, this article is based on the coal trash sorting robot platform developed by the team and the team’s early-stage coal gangue detection method research to achieve the detection of a variety of coal trash targets including gangue, aiming to solve the problem of coal trash detection in complex environments. The coal trash sorting robot provides trash detection information. The main research contents are as follows:

(1) According to the operating environment of the coal debris sorting robot, analyze the selection of the camera and light source in the visual inspection module, and build a coal debris visual inspection platform. Aiming at the problem of the lack of samples of coal sundries, web crawlers, coal preparation plant collection and laboratory simulation production methods are used to obtain coal sundries samples, and the coal sundries data sets under complex conditions are constructed through image preprocessing.

(2) In order to realize the visual detection of small samples of coal debris, the SSD target detection model is migrated and trained based on the migration learning method. Construct a coal debris detection model based on the SSD architecture according to the requirements of coal debris sorting robots. Perform performance comparison experiments with the Faster-RCNN detection model. Combine the coal debris detection accuracy and efficiency balance requirements to verify the feasibility of the detection model in this paper. Sex and reliability.

(3) In order to achieve a better balance between detection accuracy and detection speed of the SSD coal debris detection model, and at the same time solve the problems of the target scale diversity of coal debris, this paper improves the SSD coal debris detection model and constructs a lightweight coal debris detection model. First, use MobileNetV1 and MobileNetV2 feature extraction networks to replace the VGG16 network structure in the original SSD and conduct a comparative test; then use the non-maximum suppression optimization method Soft-NMS to improve the overlapping and missing detection problems of coal debris and debris detection frames; finally, use MobileNetV2 The unique reverse residual structure optimizes the model pruning, which further improves the model detection efficiency while ensuring a small loss of model accuracy.

(4) Design a coal sundries visual inspection system to realize online detection of coal and sundries targets on the belt conveyor. The system has four functions of image acquisition, preprocessing, sundries detection and visualization. Subsequently, 550 coal debris targets were used for static environmental testing and dynamic experimental testing. The experimental results show that the visual inspection system meets the needs of robot sorting and has certain practicability.

[1]王国法, 赵国瑞, 任怀伟. 智慧煤矿与智能化开采关键核心技术分析[J]. 煤炭学报, 2019,44(01):34-41.

[2]王金华, 谢和平, 刘见中, 等. 煤炭近零生态环境影响开发利用理论和技术构想[J]. 煤炭学报, 2018,43(05):1198-1209.

[3]邱硕涵, 谭章禄. 煤炭企业智慧矿山建设指标体系研究[J]. 煤炭科学技术, 2019,47(10):259-266.

[4]王国法, 张德生. 煤炭智能化综采技术创新实践与发展展望[J]. 中国矿业大学学报, 2018,47(03):459-467.

[5]王国法, 杜毅博. 智慧煤矿与智能化开采技术的发展方向[J]. 煤炭科学技术, 2019,47(01):1-10.

[6]MA Dan, DUAN Hongyu, LIU Jiangfeng. The role of gangue on the mitigation of mining-induced hazards and environmental pollution:An experimental investigation[J]. Science of the Total Environment, 2019,664:436-448.

[7]王国法, 王虹, 任怀伟, 等. 智慧煤矿2025情景目标和发展路径[J]. 煤炭学报, 2018,43(02):295-305.

[8]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.

[9]李宁. 煤矸分拣机器人控制系统研究[D]. 西安:西安科技大学,2019.

[10]王鹏, 曹现刚, 夏晶, 等. 基于机器视觉的多机械臂煤矸石分拣机器人系统研究[J]. 工矿自动化, 2019,45(09):47-53.

[11]曹现刚, 吴旭东, 王鹏, 等. 面向煤矸分拣机器人的多机械臂协同策略[J]. 煤炭学报, 2019,44(S2):763-774.

[12]Hong H , Zheng L , Zhu J , et al. Automatic Recognition of Coal and Gangue based on Convolution Neural Network[J]. 2017.

[13]苏玲玲. 基于深度学习的煤矸石识别方法研究[D]. 西安:西安科技大学,2018.

[14]Gang C, Ling S , Yan W , et al. Research on Coal Gangue Identification by Using Convolutional Nerual Network, 2018 2nd IEEE Advanced Information Management, Communicates, Electronic and Automation Control Conference (IMCEC), Xi’an, 2018.

[15]薛祯也. 基于深度学习的煤矸石识别方法研究[D]. 西安:西安科技大学,2019.

[16]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.

[17]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.

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

[19]师永春. 煤炭杂物情况及除杂工作分析[J]. 科技情报开发与经济, 2003(10):166-167.

[20]谭安君. 防止杂物进入原煤仓[J]. 科技资讯, 2013(21):120.

[21]刘京领. 在运煤皮带机上清除煤中非铁器杂物的新方法[J]. 能源技术与管理, 2005(02):41-87.

[22]张俊峰. 胶带输送机上清除煤中杂物的方法[J]. 中州煤炭, 2005(05):47-48.

[23]Zheng Gang, Jia Wei, Liu Hai Zeng, et al. Research on a Coal Gangue Identification Device[J]. Applied Mechanics & Materials, 2013,325-326:699-702.

[24]纪贵军. 火力发电厂煤中杂物分离机构的设计方法研究[D]. 杭州:浙江大学,2003.

[25]协庄煤矿创新设计应用选煤工艺水逆止杂物清除装置[J]. 煤炭加工与综合利用,2017 (05):43.

[26]方鹏. 浅槽分选机在除杂系统中的应用[J]. 煤矿现代化, 2020(01):128-129+132.

[27]杨春宇. 关于吕家坨矿洗煤厂块精煤机选除杂工艺的研究[J]. 中国资源综合利用, 2019,37(09):184-186.

[28]张军. 选煤厂杂物管控探讨[J]. 内蒙古煤炭经济, 2019(12):87-88+94.

[29]何永彬, 母海龙, 田娜, 等. 补连塔选煤厂杂物管控的措施[J]. 陕西煤炭, 2019,38(S1):146-148.

[30]王卫东, 张康辉, 吕子奇, 等. 基于机器视觉的煤中杂物智能分选系统研究[J]. 选煤 技术, 2020(02):87-91.

[31]梁蒙蒙, 周涛, 张飞飞, 等. 卷积神经网络及其在医学图像分析中的应用研究[J]. 生 物医学工程学杂志, 2018,35(06):977-985.

[32]徐信罗, 陶欢, 李存军, 等. 基于Faster R-CNN的松材线虫病受害木识别与定位[J]. 农业机械学报, 2020,51(07):228-236.

[33]李广帅, 苏娟, 李义红. 基于改进Faster R-CNN的SAR图像飞机检测算法[J]. 北京航 空航天大学学报, 2021,47(01):159-168.

[34]吕政阳, 邓涛, 张丽艳. 一种基于机器视觉的飞机钣金件跨粒度识别方法[J]. 仪器仪 表学报, 2020,41(02):195-204.

[35]姚红革, 王诚, 喻钧, 等. 复杂卫星图像中的小目标船舶识别[J]. 遥感学报, 2020,24(02):116-125.

[36]Felzenszwalb P F, Girshick R B, McAllester D, et al. Object detection with discriminatively trained part-based models[J]. IEEE transactions on pattern analysis and machine intelligence, 2009, 32(9): 1627-1645.

[37]吴帅, 徐勇, 赵东宁. 基于深度卷积网络的目标检测综述[J]. 模式识别与人工智能, 2018,31(04):335-346.

[38]Zou Z, Shi Z, Guo Y, et al. Object Detection in 20 Years: A Survey[J]. arXiv preprint arXiv:1905.05055, 2019.

[39]LeCun Y, Bengio Y, Hinton G. Deep learning[J]. nature, 2015, 521(7553): 436-444.

[40]Goodfellow I, Bengio Y, Courville A, et al. Deep learning[M]. Cambridge:MIT press, 2016.

[41]Girshick R, Donahue J, Darrell T, et al. Rich feature hierarchies for accurate object detection and semantic segmentation [C]. 2014 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), June 23-28, 2014, Columbus, OH, USA. New Jersey, IEEE Press, 2014: 580-587.

[42]Girshick R. Fast R-CNN [C]. the IEEE International Conference on Computer Vision, December 7-13, 2015, Chile. New Jersey, IEEE Press, 2015: 1440-1448.

[43]Ren S Q, He K M, Girshick R, et al. Faster R-CNN: towards real-time object detection with region proposal networks [J]. Computer Science, 2015(99): 1.

[44]He K, Zhang X, Ren S, et al. Spatial pyramid pooling in deep convolutional networks for visual recognition[J]. IEEE transactions on pattern analysis and machine intelligence, 2015, 37(9): 1904-1916.

[45]Lin T Y, Dollár P, Girshick R, et al. Feature pyramid networks for object detection[C]//Proceedings of the IEEE conference on computer vision and pattern recognition. 2017: 2117-2125.

[46]Redmon J, Divvala S, Girshick R, Farhadi A. You only look once: unified, real-time object detection[C]. 2016 IEEE Conference on Computer Vision and Pattern Recognition 2016: 779-788.

[47]Liu W, Anguelov D, Erhan D, Szegedy C, Reed S, Fu C Y, Berg A C. SSD: single shot multibox detector[M]. Computer Vision-ECCV 2016, Springer International Publishing, 2016, 9905: 21-37.

[48]Lin T Y, Goyal P, Girshick R, et al. Focal loss for dense object detection[C]//Proceedings of the IEEE international conference on computer vision. 2017: 2980-2988.

[49]李旭冬, 叶茂, 李涛. 基于卷积神经网络的目标检测研究综述[J]. 计算机应用研究, 2017,34(10):2881-2886+2891.

[50]余永维, 韩鑫, 杜柳青. 基于Inception-SSD算法的零件识别[J]. 光学精密工程, 2020,28(08):1799-1809.

[51]戴文君, 常天庆, 张雷, 等. 面向坦克火控系统的多尺度形变目标检测方法[J]. 计算 机辅助设计与图形学学报, 2019,31(12):2082-2090.

[52]王纪武, 罗海保, 鱼鹏飞, 等. 基于Faster R-CNN的无人机航拍图像小目标检测(英 文)[J]. Journal of Measurement Science and Instrumentation, 2020,11(01):11-16.

[53]朱兴动, 田少兵, 黄葵, 等. 基于深度卷积神经网络的舰载机目标检测[J]. 计算机应 用, 2020,40(05):1529-1533.

[54]赵望宇, 李必军, 单云霄, 等. 融合毫米波雷达与单目视觉的前车检测与跟踪[J]. 武 汉大学学报(信息科学版), 2019,44(12):1832-1840.

[55]卫星, 张海涛, 陆阳, 等. 结合并行特征传递深度学习网络的矿井行人检测[J]. 计算 机辅助设计与图形学学报, 2019,31(12):2091-2100.

[56]李鹏飞. 基于卷积神经网络的煤堆烟雾检测算法研究[D]. 秦皇岛:燕山大学, 2020.

[57]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.

[58]郜亚松. 基于深度学习的煤矸石在线识别定位研究与设计[D]. 邯郸:河北工程大 学,2020.

[59]Sun R. Optimization for deep learning: theory and algorithms[J]. arXiv preprint arXiv:1912.08957, 2019.

[60]李利民. SAR图像目标检测的并行处理研究[D]. 哈尔滨:哈尔滨工业大学, 2010.

[61]Garcia M A, Munteanu E, Peana S. Method and system for LCD panel light source selection: U.S. Patent 6,943,770[P]. 2005-9-13.

[62]Van de Ven A P, Negley G H. Lighting device and lighting method: U.S Patent 7,213,940[P]. 2007-5-8.

[63]Tan C, Sun F, Kong T, et al. A survey on deep transfer learning[C]//International conference on artificial neural networks. Springer, Cham, 2018: 270-279.

[64]Pan S J, Yang Q. A survey on transfer learning[J]. IEEE Transactions on knowledge and data engineering, 2009, 22(10): 1345-1359.

[65]王晋东. 迁移学习简明手册[Z]. https://zhuanlan.zhihu.com/p/35352154

[66]臧文华. 基于生成对抗网络的迁移学习算法研究[D]. 成都:电子科技大学,2018.

[67]翟杰. 基于部分知识的迁移学习方法研究[D]. 南京:南京邮电大学,2019.

[68]Yosinski J, Clune J, Bengio Y, et al. How transferable are features in deep neural networks[J]. arXiv preprint arXiv:1411.1792, 2014.

[69]Howard A G, Zhu M, Chen B, et al. Mobilenets: Efficient convolutional neural networks for mobile vision applications[J]. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2017.

[70]Howard A, Zhmoginov A, Chen L C, et al. Inverted residuals and linear bottlenecks: Mobile networks for classification, detection and segmentation[J]. 2018.

[71]He K, Zhang X, Ren S, et al. Deep residual learning for image recognition[C]//Pr- oceedings of the IEEE conference on computer vision and pattern recognition. 201 6: 770-778.

[72]Bodla N, Singh B, Chellappa R, et al. Soft-NMS-improving object detection with one line of code[C]//Proceedings of the IEEE international conference on computer vision. 2017: 5561-5569.