煤炭是国家的重要能源之一，国家一直以来不断加强对能源的开采挖掘领域的探索。近年来，国家针对煤炭行业颁布了新的政策，大力倡导煤炭的清洁使用的同时做好“碳达峰、碳中和”工作。实现煤矸石的绿色开采，离不开煤炭的分选工作，传统的分选技术主要分为干选法和湿选法两大类，主要的弊端在于生产的智能化程度不高且成本过大，造成资源的浪费。随着人工智能的不断发展，煤炭分选技术朝着智能化的方向迈进成为必然的趋势，论文提出用深度学习的方法解决煤矸石分选智能化问题，研究内容分为以下几点：

（1）由于煤矸石数据集属于非公开数据集数量有限，需要对其进行扩充。传统的平移、旋转等图像增强算法，只是使煤矸石的数量增加并没有改变煤矸石形状的多样性，无法适应结构较深的网络。采用改进的ConsinGAN网络丰富煤矸石的多样性，并用相关的指标衡量生成图像与真实图像的差异进行对比实验。结果表明，改进的ConsinGAN网络可以生成更加多样的煤矸石图片，使煤矸石特征多样化。

（2）皮带中传输的煤矸石大多存在堆叠、粘连的现象，为了准确获取煤矸石的形状位置信息，提出MAS-Unet分割网络，在特征提取部分加入多条空洞率的深度卷积支路模块（Multi-Rate Dilated Convolution Block，MDCB），减少部分计算量的同时增加信息提取的能力，同时为了减少经过跳跃连接后特征信息损失，在MAS-Unet中将跳跃连接替换为ASPP模块，融合编码器的不同特征获得多种尺度的信息，并且在解码器部分加入SEC模块，给与每个通道不同的关注程度，从而重点关注煤矸石信息，忽略掉无关信息。实验部分验证了MAS-Unet在煤矸石堆叠、粘连情况下均能准确进行分割，达到95.08%的准确率，在不同光照下，MAS-Unet具有一定的鲁棒性，可以适应不同的光照条件。

（3）为了更加精准实时的对煤矸石进行检测，对YOLOv5网络进行改进，用Ghost卷积代替原来的普通卷积对特征进行去冗余处理，减少重复的计算。煤矸石轮廓等关键信息是网络学习的重点需要赋予较高的权重，在Backbone和Neck中采用CBAM注意力机制关注不同尺度的空间和位置信息，提高煤矸石的识别率，并且预测头中加入针对煤矸石小颗粒的识别层，尽可能多的检测到所有目标。最后，将改进的YOLOv5与MAS-Unet网络结合，使得融合后网络不仅可以检测到目标同时能的到煤矸石的形状等信息。实验部分通过消融实验和对比实验对改进的YOLOv5网络性能进行验证，改进后网络的精度比原模型提高3.8%，模型大小降低了37.5%，单张煤矸石图片训练时间缩短了7ms以及融合后的网络对煤矸石样本进行检测有较高的检测结果。

Coal is one of the country's important energy, the country has been constantly strengthening the exploration of energy mining and mining fields..In recent years, China has promulgated new policies for the coal industry, vigorously advocating the clean use of coal while doing a good job of "Carbon peaking, Carbon neutral".The realization of green mining of coal gangue is inseparable from the separation of coal. Traditional separation methods are mainly divided into two categories: dry separation and wet separation. The main drawback is that the production is not intelligent and the cost is too high, resulting in the waste of resources.With the continuous development of artificial intelligence, it has become an bound to tendency for coal separation technology to move towards the direction of intelligence. This paper proposes a deep learning method to solve the intelligent problem of coal gangue separation. The research content is divided into the following points:

(1)Because the coal gangue data set is a non-public data set with limited quantity, it needs to be expanded.Traditional image enhancement algorithms such as translation and rotation only increase the number of coal gangue, but do not change the diversity of the shape of coal gangue, and cannot adapt to the network with deep structure. The improved ConsinGAN network is used to enrich the diversity of coal gangue, Using relevant indicators to estimate the difference between the generated image and the original image, a comparative experiment was conducted.The results show that the improved ConsinGAN network can generate more diverse and realistic images of coal gangue, and make the characteristics of coal gangue diverse.

(2)Most of the coal gangue transported in the belt has the phenomenon of stacking, adhesion and multi-scale. In order to accurately obtain the shape and position information of the coal gangue, MAS-Unet segmentation network is proposed. In the feature extraction part, a number of deep convolution branch modules with void ratio are added to reduce part of the calculation and increase the ability of information extraction.In order to reduce the loss of feature information obtained by jump connection, ASPP module is added to MAS-Unet to fuse the encoder to obtain information of different scales, and SEC module is added to the decoder to give different attention to each channel, So as to focus on coal gangue information and ignore irrelevant information.In the experimental part, MAS-Unet can accurately segment the coal gangue under the conditions of stacking and adhesion , with the accuracy of 95.08%. MAS-Unet has certain robustness under different illumination.

(3)In order to detect coal gangue more accurately and in real-time, the YOLOv5 network has been improved by using Ghost convolution instead of the original ordinary convolution to remove redundant features and reduce repetitive calculations.The key information such as the contour of coal gangue is the focus of network learning, which needs to be given a higher weight. CBAM attention mechanism is used in Backbone and Neck to focus on the space and location information of different scales to improve the recognition rate of coal gangue, and the recognition layer for small particles of coal gangue is added in the prediction head to detect all targets as much as possible.Finally, The improved YOLOv5 and combined with the MAS-Unet network, so that the fused network can not only detect the target but also obtain information such as the shape of coal gangue. The experimental part verified the performance of the improved YOLOv5 network through ablation experiments and comparative experiments. The accuracy of the improved network was increased by 3.8% compared to the original model, the model size was reduced by 37.5%, the training time for a single coal gangue image was shortened by 7 ms, and the fused network had high detection results for coal gangue samples.

[1]Du L, Dinçer H, Ersin İ, et al. IT2 fuzzy-based multidimensional evaluation of coal energy for sustainable economic development[J]. Energies, 2020, 13(10): 2453.

[2]Chen J, Li Z, Song M, et al. Economic and intensity effects of coal consumption in China[J]. Journal of Environmental Management, 2022, 301: 113912.

[3]赵开功, 李彦平. 我国煤炭资源安全现状分析及发展研究[J]. 煤炭工程, 2018, 10: 185-189.

[4]李静, 温鹏飞, 何振嘉. 煤矸石的危害性及综合利用的研究进展[J]. 煤矿机械, 2017, 38(11): 128-130.

[5]刘存玉. “双碳”形势下的煤矿绿色开采探索[J]. 中国煤炭工业, 2022.

[6]刘富. 露天煤矿绿色开采技术分析[J]. 工程学研究与实用, 2023, 4(5): 128-130.

[7]杨玉龙. 煤矸石固体废物对环境的危害及防治对策[J]. 云南化工, 2022.

[8]董增澳, 李萍, 贾一雪, 等. 煤矸石综合利用与资源化处理研究进展[J]. 环境保护前沿, 2021, 11(2): 363-371.

[9]王晓栋, 张玥, 陈松, 等. 煤矸石资源化利用的研究进展[J]. 化学工程师, 2021, 35(4): 68-69.

[10] 陈强. 论我国选煤技术发展现状及对煤炭清洁利用的影响[J]. 工程建设(2630-5283), 2022(008):005.

[11] 杨林青, 胡方坤. 我国选煤技术的现状及发展[J]. 煤炭技术, 2010, 29(5): 109-111.

[12]王梅. 干法选煤技术的应用现状及发展趋势[J]. 山东工业技术, 2018(21):1.

[13]李思维, 常博, 刘昆轮, 等. 煤炭干法分选的发展与挑战[J]. Clean Coal Technology, 2021, 27(5).

[14]Zhou C, Liu X, Zhao Y, et al. Recent progress and potential challenges in coal upgrading via gravity dry separation technologies[J]. Fuel, 2021, 305: 121430.

[15]Jaiswal S, Tripathy S K, Banerjee P K. An overview of reverse flotation process for coal[J]. International Journal of Mineral Processing, 2015, 134: 97-110.

[16]Zhao Y, Wang S, Cheng G, et al. Study on coal and gangue recognition method based on the combination of X-ray transmission and diffraction principle[J]. Energy Sources, Part A: Recovery, Utilization, and Environmental Effects, 2022, 44(4): 9716-9728.

[17]Gui X, Liu J, Cao Y, et al. Coal preparation technology: Status and development in China[J]. Energy & Environment, 2015, 26(6-7): 997-1013.

[18]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]. Coal Science & Technology (0253-2336), 2022, 50(8).

[19]吴景涛. 机器视觉技术在煤矸石智能识别中的应用研究[D]. 中国矿业大学, 2019.

[20]Prasad Choudhary D. Cometary and Solar Observations with Small Telescopes Connected to a Computer Network[C]//Selected Papers from Activities held in 2000. Seminars of the United Nations Programme on Space Applications, Volume 12. United Nations, New York, 2001., p. 87. 2001, 12: 87.

[21]Tripathy D P, Guru Raghavendra Reddy 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: 109-117.

[22]Tripathi S, Singh L, Arora H. Face recognition machine vision system using eigenfaces[J]. International Journal of Recent Trends in Engineering, 2009, 2(2): 1.

[23]Hong H, Zheng L, Zhu J, et al. Automatic recognition of coal and gangue based on convolution neural network[J]. arXiv preprint arXiv:1712.00720, 2017.

[24]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.

[25]Liu Q, Li J, Li Y, et al. Recognition methods for coal and coal gangue based on deep learning[J]. IEEE Access, 2021, 9: 77599-77610.

[26]Alfarzaeai M S, Niu Q, Zhao J, et al. Coal/gangue recognition using convolutional neural networks and thermal images[J]. IEEE Access, 2020, 8: 76780-76789.

[27]Lv Z, Wang W, Xu Z, et al. Cascade network for detection of coal and gangue in the production context[J]. Powder Technology, 2021, 377: 361-371.

[28]Lai W, Zhou M, Hu F, et al. A study of multispectral technology and two-dimension autoencoder for coal and gangue recognition[J]. IEEE Access, 2020, 8: 61834-61843.

[29]来文豪, 周孟然, 王锦国, 等. 多光谱波段筛选的煤矸石快速定位[J]. 中国激光, 2021, 48(16): 1611001.

[30]王锐, 桂志国, 刘祎, 等. 基于 X 射线和结构光相机的煤矸石分选方法研究[J]. 中北大学学报: 自然科学版, 2021, 42(2): 123-128.

[31]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(3): 1117-1121.

[32]陈立, 杜文华, 曾志强, 等. 基于小波变换的煤矸石自动分选方法[J]. 工矿自动化, 2018, 44(12): 60-64.

[33]耿秀云. 基于 X 光图像处理的煤矸石自动分选系统的研究[D]. 沈阳: 东北大学 2014: 30-34, 2014.

[34]马宏伟, 孙那新, 张烨, 等. 煤矸石分选机器人动态目标稳定抓取轨迹规划[J]. 工矿自动化, 2022, 48(4): 20-30.

[35]朱子祺, 李创业, 代伟. 基于 G-RRT~* 算法的煤矸石分选机器人路径规划[J]. 工矿自动化, 2022.

[36]杜振国. 卷积神经网络模型压缩方法研究与应用[D]. 南昌大学, 2021.

[37]Krizhevsky A, Sutskever I, Hinton G E. ImageNet Classification with Deep Convolutional Neural Networks (AlexNet) ImageNet Classification with Deep Convolutional Neural Networks (AlexNet) ImageNet Classification with Deep Convolutional Neural Networks[J].

[38]Simonyan K, Zisserman A. Very deep convolutional networks for large-scale image recognition[J]. arXiv preprint arXiv:1409.1556, 2014.

[39]Szegedy C ,Liu W ,Jia Y , et al. Going Deeper with Convolutions[J]. IEEE Computer Society, 2014.

[40]He K ,Zhang X ,Ren S , et al. Deep Residual Learning for Image Recognition[J]. IEEE, 2016.

[41]Huang G, Liu Z, Laurens V, et al. Densely Connected Convolutional Networks[J]. IEEE Computer Society, 2016.

[42]Long J, Shelhamer E, DarrellT. Fully convolutional networks for semantic segmentation[C]//Proceedings of the IEEE conference on computer vision and pattern recognition. 2015: 3431-3440.

[43]Redmon J, Divvala S, Girshick R, et al. You only look once: Unified, real-time object detection[C]//Proceedings of the IEEE conference on computer vision and pattern recognition. 2016: 779-788.

[44]Wei L , Dragomir A ,Dumitru E , et al. SSD: Single Shot MultiBox Detector[J]. Springer, Cham, 2016.

[45]Ronneberger O,Fischer P,Brox T. U-net: Convolutional networks for biomedical image segmentation[C]//Medical Image Computing and Computer-Assisted Intervention–MICCAI 2015: 18th International Conference, Munich, Germany, October 5-9, 2015, Proceedings, Part III 18. Springer International Publishing, 2015: 234-241.

[46]Goodfellow I ,Pouget-Abadie J ,Mirza M , et al. Generative Adversarial Nets[C]// Neural Information Processing Systems. MIT Press, 2014.

[47]王坤峰, 苟超, 段艳杰, 等. 生成式对抗网络 GAN 的研究进展与展望[J]. 自动化学报, 2017, 43(3): 321-332.

[48]Wang P, Chen P, Yuan Y, et al. Understanding convolution for semantic segmentation[C]//2018 IEEE winter conference on applications of computer vision (WACV). Ieee, 2018: 1451-1460.

[49]Zhao H, Shi J, Qi X, et al. Pyramid scene parsing network[C]//Proceedings of the IEEE conference on computer vision and pattern recognition. 2017: 2881-2890.

[50]Badrinarayanan V, Kendall A, Cipolla R. Segnet: A deep convolutional encoder-decoder architecture for image segmentation[J]. IEEE transactions on pattern analysis and machine intelligence, 2017, 39(12): 2481-2495.

[51]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.

[52]Girshick R, Donahue J, Darrell T, et al. Rich feature hierarchies for accurate object detection and semantic segmentation[C]//Proceedings of the IEEE conference on computer vision and pattern recognition. 2014: 580-587.

[53]Girshick R . Fast R-CNN[J]. Computer Science, 2015.

[54]Ren S ,He K ,Girshick R , et al. Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks[J]. IEEE Transactions on Pattern Analysis & Machine Intelligence, 2017, 39(6):1137-1149.

[55]Liu S, Qi L, Qin H, et al. Path aggregation network for instance segmentation[C]//Proceedings of the IEEE conference on computer vision and pattern recognition. 2018: 8759-8768.

[56]Duan K, Bai S, Xie L, et al. Centernet: Keypoint triplets for object detection[C]//Proceedings of the IEEE/CVF international conference on computer vision. 2019: 6569-6578.