带式输送机是矿井运输系统的关键设备。由于井下开采环境复杂，在皮带运输过程中经常会出现大量异物，容易导致皮带撕裂、转接处堵塞等异常情况。然而，现有的方法中人工检测效率低，安全风险高；雷达探测成本高，维护困难；金属探测器使用难度大，检测类别少。针对上述问题，论文对基于深度学习的异物识别技术展开研究，设计在嵌入式平台上具有高精度，能实时边缘计算的带式输送机异物识别系统，并在工业现场进行实验和性能分析。主要的工作如下：

（1）针对现有目标检测算法对异物特征提取效率不高，容易出现误检现象等问题，提出一种基于YOLOv5的异物识别算法。首先采集矿石流异物数据，利用基于背景分割的数据增强方法将样本扩充后进行标注，构建异物数据集；研究基于CBAM注意力机制和Ghost模块的主干网络、加权双向特征金字塔的特征融合网络和基于Focal Loss的损失函数，提高模型在复杂背景中的特征提取效率；再利用K-means算法初始化锚框参数，训练异物识别模型YOLOv5l_GC，在异物数据集上验证改进算法的有效性。

（2）针对上述模型存在网络参数量较大，难以在嵌入式平台部署等问题，研究多层网络通道剪枝算法以及知识蒸馏算法。利用多层网络通道剪枝算法对YOLOv5l_GC进行裁剪，剔除模型冗余参数；研究知识蒸馏算法，弥补因模型压缩带来的精度损失；最后在NVIDIA Jetson Xavier NX嵌入式平台上验证模型压缩算法的有效性。

（3）根据工业现场实际需求，研发一套带式输送机异物识别系统，通过对监控视频流进行处理分析，实时识别皮带异物并发送报警信号。结果表明，该系统实现了在嵌入式平台上高精度的实时边缘计算，对矿井运输系统的安全稳定运行具有重要意义。

Belt conveyor is the key equipment of mine transportation system. Due to the complex underground mining environment, a large number of foreign objects often appear in the process of belt transportation, which may easily lead to abnormal conditions such as belt tearing and blockage of the transfer. However, in the existing methods, the manual detection efficiency is low, and the safety hazard is large; the radar detection cost is high, and the maintenance is difficult; the metal detector is difficult to deploy and has few detection categories. In response to the above problems, this paper studied the foreign object recognition technology based on deep learning, designed a belt conveyor foreign object recognition system with high precision and real-time edge computing on an embedded platform, and conducted experiments and performance analysis on the industrial site. The main work of this paper is as follows:

(1) Aiming at the problems that the existing target detection algorithm is not efficient for foreign object feature extraction and prone to false detection, a foreign object recognition algorithm based on YOLOv5 is proposed. First, the foreign matter data of ore flow is collected, and the samples are expanded and labeled by the data enhancement method based on background segmentation to construct a foreign matter data set. Then, the backbone network based on CBAM attention mechanism and Ghost module, the feature fusion network based on weighted bidirectional feature pyramid and the loss function based on Focal Loss are studied to improve the feature extraction efficiency of the model in complex backgrounds. Finally, K-means algorithm is used to initialize anchors box parameters, the foreign object recognition model YOLOv5l_GC is trained, and the effectiveness of the improved algorithm is verified on the foreign object data set.

(2) In view of the problems that the above models have a large amount of parameters and are difficult to deploy on embedded platforms, network pruning and knowledge distillation algorithms are studied. First, the multi-layer network channel pruning algorithm is used to prune YOLOv5l_GC, and the redundant parameters of the model are eliminated. Then, the knowledge distillation algorithm is studied to make up for the loss of accuracy caused by model compression. Finally, the effectiveness of the model compression algorithm is verified on the NVIDIA Jetson Xavier NX embedded platform.

(3) According to the actual needs of the industrial site, a set of belt conveyor foreign body identification system has been developed. By processing and analyzing the monitoring video stream, it can identify the belt foreign body in real time and send an alarm signal. The results show that the system realized high-precision real-time edge computing on the embedded platform, which is of great significance to the safe and stable operation of the mine transportation system.

[1] Cai M, Li P, Tan W, et al. Key Engineering Technologies to Achieve Green, Intelligent, and Sustainable Development of Deep Metal Mines in China[J]. Engineering, 2021, 7(11): 1513-1517.

[2] 吕云龙. 2021年钢铁行业价格形势分析与2022年展望[J]. 中国物价, 2022, (02): 19-20+69.

[3] 苗长云, 邵琦. 基于声音的带式输送机输送带纵向撕裂检测方法[J]. 天津工业大学学报, 2021, 40(06): 70-75+82.

[4] 高瑞, 苗长云, 苗笛, 李现国. 输送带故障检测多视点图像自适应增强方法[J]. 煤炭学报, 2017, 42(S2): 594-602.

[5] 李轩. 工信部等:联合印发《关于促进钢铁工业高质量发展的指导意见》[J]. 中国设备工程, 2022, (04): 1.

[6] Jiao L, Zhang F, Liu F, et al. A survey of deep learning-based object detection[J]. IEEE access, 2019, 7(01): 128837-128868.

[7] Wang Y, Han Q, Li Y, et al. Video Smoke Detection Based on Multi-feature Fusion and Modified Random Forest[J]. Engineering Letters, 2021, 29(03): 1115-1112.

[8] Ren J, Wang Y. Overview of Object Detection Algorithms Using Convolutional Neural Networks[J]. Journal of Computer and Communications, 2022, 10(01): 115-132.

[9] Wu S, Zhong S, Liu Y. Deep Residual Learning for Image Steganalysis[J]. Multimedia tools and Applications, 2018, 77(9): 10437-10453.

[10] Li Z, Sun Y, Tian G, et al. A Compression Pipeline for One-stage Object Detection Model[J]. Journal of Real-Time Image Processing, 2021, 18(6): 1949-1962.

[11] 王军, 冯孙铖, 程勇. 深度学习的轻量化神经网络结构研究综述[J]. 计算机工程, 2021, 47(08): 1-13.

[12] Chen J, Ran X. Deep Learning with Edge Computing: A Review[J]. Proceedings of the IEEE, 2019, 107(08): 1655-1674.

[13] 郭永存, 何磊, 刘普壮, 王希. 煤矸双能X射线图像多维度分析识别方法[J]. 煤炭学报, 2021, 46(01): 300-309.

[14] 王燕, 郭潇樯, 刘新华. 带式输送机大块异物视觉检测系统设计[J]. 机械科学与技术, 2021, 40(12): 1939-1943.

[15] 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.

[16] 郝乐. 地面毛煤煤矸石目标检测算法研究[D]. 西安: 西安科技大学, 2021.

[17] 吕志强. 复杂环境下煤矿皮带运输异物图像识别研究[D]. 徐州: 中国矿业大学, 2020.

[18] Wang Y, Wang Y, Dang L. Video Detection of Foreign Objects on the Surface of Belt Conveyor Underground Coal Mine Based on Improved SSD[J]. Journal of Ambient Intelligence and Humanized Computing, 2020, 9(06): 1-10.

[19] 胡璟皓, 高妍, 张红娟, 靳宝全. 基于深度学习的带式输送机非煤异物识别方法[J]. 工矿自动化, 2021, 47(06): 57-62+90.

[20] 卢才武, 闫雪颂, 刘力, 何旭乾, 樊腊梅. 一种改进的无锚框式金属矿带式输送机异物检测方法[J]. 采矿技术, 2022, 22(01): 150-154+162.

[21] 许鹏. 基于边缘计算的煤矿井下皮带异物检测关键技术研究[D]. 徐州: 中国矿业大学, 2021.

[22] Krizhevsky A, Sutskever I, Hinton G E. Imagenet Classification with Deep Convolutional Neural Networks[J]. Communications of the ACM, 2017, 60(06): 84-90.

[23] 谢富, 朱定局. 深度学习目标检测方法综述[J]. 计算机系统应用, 2022, 31(02): 1-12.

[24] Bharati P, Pramanik A. Deep learning techniques—R-CNN to Mask R-CNN: A Survey[J]. Computational Intelligence in Pattern Recognition, 2020, 18(08): 657-668.

[25] Li J, Liang X, Shen S M, et al. Scale-aware Fast R-CNN for Pedestrian Detection[J]. IEEE transactions on Multimedia, 2017, 20(04): 985-996.

[26] 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 and Machine Intelligence, 2016, 39(06): 1137-1149.

[27] Zhang H, Wang Y, Dayoub F, et al. VarifocalNet: An IoU-aware Dense Object Detector[C]// Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. IEEE, 2021: 8514-8523.

[28] Nguyen D T, Nguyen T N, Kim H, et al. A High-throughput and Power-efficient FPGA Implementation of YOLO CNN for Object Detection[J]. IEEE Transactions on Very Large Scale Integration (VLSI) Systems, 2019, 27(8): 1861-1873.

[29] Kumar A, Srivastava S. Object Detection System Based on Convolution Neural Networks Using Single Shot Multi-box Detector[J]. Procedia Computer Science, 2020, 171: 2610-2617.

[30] Wang L, Yang S, Yang S, et al. Automatic Thyroid Nodule Recognition and Diagnosis in Ultrasound Imaging with The YOLOv2 Neural Network[J]. World journal of surgical oncology, 2019, 17(1): 1-9.

[31] 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. IEEE, 2017: 2980-2988.

[32] Zhao L, Li S. Object Detection Algorithm Based on Improved YOLOv3[J]. Electronics, 2020, 9(3): 537.

[33] Tan M, Pang R, Le Q V. Efficientdet: Scalable and efficient object detection[C]// Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. IEEE, 2020: 10781-10790.

[34] Abdurahman F, Fante K A, Aliy M. Malaria Parasite Detection in Thick Blood Smear Microscopic Images Using Modified YOLOV3 and YOLOV4 Models[J]. BMC Bioinformatics, 2021, 22(01): 1-17.

[35] 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.

[36] Zhang Q Q, Zhu Z J, Ge Z F, et al. Key Components Detection and Identification of Transmission Lines Based on An Improved CornerNet Network[J]. Journal of Computers, 2021, 32(2): 124-136.

[37] 张翼翔, 林松, 李雪. 基于CenterNet-GhostNet的选煤厂危险区域人员检测[J]. 工矿自动化, 2022, 48(04): 66-71.

[38] Chen Y, Zhang Z, Cao Y, et al. Reppoints v2: Verification Meets Regression for Object Detection[J]. Advances in Neural Information Processing Systems, 2020, 33: 5621-5631.

[39] Song J, Zhao Y, Song W, et al. Fisheye Image Detection of Trees Using Improved YOLOX for Tree Height Estimation[J]. Sensors, 2022, 22(10): 3636.

[40] 黄靖淞, 左颢睿, 张建林. 轻量化目标检测算法研究及应用[J]. 计算机工程: 2020, 45(12): 1-8.

[41] Kim W, Jung W S, Choi H K. Lightweight Driver Monitoring System Based on Multi-task Mobilenets[J]. Sensors, 2019, 19(14): 3200.

[42] Lee H J, Ullah I, Wan W, et al. Real-time Vehicle Make and Model Recognition with The Residual SqueezeNet Architecture[J]. Sensors, 2019, 19(05): 982.

[43] Yin J, Guo L, Jiang W, et al. ShuffleNet-inspired Lightweight Neural Network Design for Automatic Modulation Classification Methods in Ubiquitous IoT Cyber–physical Systems[J]. Computer Communications, 2021, 176: 249-257.

[44] Liu J, Wang X. Early Recognition of Tomato Gray Leaf Spot Disease Based on MobileNetv2-YOLOv3 Model[J]. Plant Methods, 2020, 16(1): 1-16.

[45] Ran H, Wen S, Wang S, et al. Memristor-based Edge Computing of ShuffleNetV2 for Image Classification[J]. IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems, 2020, 40(8): 1701-1710.

[46] Zhao L, Wang L. A new lightweight network based on MobileNetV3[J]. KSII Transactions on Internet and Information Systems (TIIS), 2022, 16(1): 1-15.

[47] Paoletti M E, Haut J M, Pereira N S, et al. GhostNet for Hyperspectral Image Classification[J]. IEEE Transactions on Geoscience and Remote Sensing, 2021, 59(12): 10378-10393.

[48] 刘月峰, 边浩东, 何滢婕, 郭威, 张小燕. 基于幅值迭代剪枝的多目标奶牛进食行为识别方法[J]. 农业机械学报, 2022, 53(02): 274-281.

[49] Hsiao T Y, Chang Y C, Chou H H, et al. Filter-based Deep-compression with Global Average Pooling for Convolutional Networks[J]. Journal of Systems Architecture, 2019, 95: 9-18.

[50] Wang Z, Li F, Shi G, et al. Network Pruning Using Sparse Learning and Genetic Algorithm[J]. Neurocomputing, 2020, 404: 247-256.

[51] Wang J, Yang L, Huo Z, et al. Multi-label Classification of Fundus Images with Efficientnet[J]. IEEE Access, 2020, 8: 212499-212508.

[52] Chen X, Cheng Y, Wang S, et al. The Elastic Lottery Ticket Hypothesis[J]. Advances in Neural Information Processing Systems, 2021, 34: 26609-26621.

[53] Hinton G, Vinyals O, Dean J. Distilling the Knowledge in a Neural Network[J]. Computer Science, 2015, 14(07): 38-39.

[54] Mehta R, Ozturk C. Object Detection at 200 Frames per Second[C]//Proceedings of the European Conference on Computer Vision (ECCV) Workshops. Springer, Munich, 2018: 8-14.

[55] Liu Y, Zhang W, Wang J. Adaptive Multi-teacher Multi-level Knowledge Distillation[J]. Neurocomputing, 2020, 415: 106-113.

[56] 张翠军, 赵娜. 基于概率神经网络改进的GrabCut算法[J]. 激光与光电子学进展, 2021, 58(02): 243-250.

[57] 郝帅, 马瑞泽, 赵新生, 安倍逸, 张旭, 马旭. 基于卷积块注意模型的YOLOv3输电线路故障检测方法[J]. 电网技术, 2021, 45(08): 2979-2987.

[58] 杜京义, 陈瑞, 郝乐, 史志芒. 煤矿带式输送机异物检测[J]. 工矿自动化, 2021, 47(08): 77-83.

[59] 李江昀, 赵义凯, 薛卓尔, 蔡铮, 李擎. 深度神经网络模型压缩综述[J]. 工程科学学报, 2019, 41(10): 1229-1239.