目前，真实可靠的煤矿井下地理信息是实现煤矿智能开采的重要前提，实时巷道数据刷新与重建是提升井下地理信息的重要组成，然而，受煤矿巷道空间狭小、照度低且不均匀、视觉受限，动态开采伴随着粉尘、水雾等复杂环境的影响，导致视觉与激光雷达等感知精度差、可靠性低、适应能力不足，难以实现高精度巷道实时表征。毫米波雷达对粉尘、水雾等复杂环境具有较高适应性，基于此，提出毫米波雷达阵列的煤矿巷道数字建模研究，对煤矿巷道进行三维扫描测量与精确重构，实时提供准确的巷道空间信息和地质特征，为煤矿智能开采、装备精确定位定向与路径规划等提供重要支持。

（1）针对煤矿巷道环境复杂、信息获取困难的问题，研究粉尘、水雾、巷道结构等环境因素对毫米波雷达信号的影响规律，搭建了模拟环境实验平台。采集不同（状态、粒径、浓度）粉尘环境与不同水雾环境下的毫米波雷达信号信息数据，建立煤矿巷道围岩毫米波信号衰减模型，对比分析3种环境因素毫米波雷达的感知机理。研究并分析不同状态下粉尘粒径与浓度、不同形状下水雾粒径与喷洒面积和理想巷道结构的毫米波雷达感知特性，结果表明不同静态粉尘粒径与浓度对毫米波雷达信号均无明显影响，但动态粉尘浓度越大、粒径越大时，对毫米波雷达信号影响越明显；在水雾状态下，水雾的粒径与喷洒面积对毫米波雷达信号影响较小可不考虑，但水滴、水柱对毫米波雷达信号影响较为明显；理想巷道围岩环境对毫米波信号的衰减较小，相对于雷达本身信号损耗可忽略不计。可见，毫米波雷达在煤矿巷道复杂环境条件下具有较高的适应性，为构建煤矿三维巷道空间模型提供高精度煤矿巷道点云数据，为巷道数字建模提供新方法。

（2）针对煤矿复杂环境下毫米波雷达阵列的巷道数字建模问题，首先，分析毫米波点云单帧数据特点，提出毫米波雷达点云多层滤波降噪处理与动态子图配准方法，滤除离群点、孤立点及虚假目标点，实现单雷达点云配准；其次，为解决单一毫米波雷达无法一次获取完整巷道的围岩信息问题，结合毫米波雷达高程视场特点，构建改进迭代近邻点（Iterative Closest Point， ICP）配准的毫米波雷达阵列感知方法，提出毫米波雷达阵列的巷道点云融合模型；最后，为解决巷道毫米波雷达点云不均匀与稀疏问题，提出点云密度加权插值的泊松表面重建方法，细化巷道围岩轮廓，贴近实际巷道形状尺寸，实现数字巷道模型的精确重构。通过实验分析，利用点云密度加权法巷道平均点云密度提升29.1%，解决了巷道模型重构时毫米波雷达阵列点云数据不均匀问题，提高了点云的平均密度。

（3）针对毫米波雷达阵列点云空间不均匀、重构模型不完整等问题，分析了注意力机制和生成对抗网络原理及其优势，提出基于注意力机制与生成对抗网络的煤矿巷道数字建模优化方法。设计双模式层次化采样单元与差异化特征扩展单元，捕获不同层次的点云特征，实现点云特征聚合。充分考虑毫米波点云密度与空间分布，提出点云密度空间注意力机制，对密度与空间位置特征进行关联，关注潜在的局部点云特征。建立基于空间注意力机制的双判别器模式，训练关注局部点云特征与全局空间信息，生成高质量巷道点云数据，提高巷道空间坐标的精确度。通过实验分析，基于注意力机制与生成对抗网络的煤矿巷道数字建模优化方法，能够提高点云密度及均匀性，实现点云稠密化与补全，重建巷道更为完善与平整。

（4）通过实验室实验测试与模拟巷道环境验证，在揭示煤矿复杂环境条件下的毫米波雷达阵列感知机理基础上，实现了基于毫米波雷达阵列的煤矿巷道数字建模，结果表明，该方法能够充分展示真实巷道围岩信息，宽度平均绝对误差百分比为0.82%，高度平均绝对误差百分比为0.72%。通过基于注意力机制与生成对抗网络的煤矿巷道数字建模优化方法，巷道重建后的宽度平均绝对误差为2.02cm，相对优化前提升28%，高度平均绝对误差为1.52cm，相对优化前提升23%，实现了煤矿巷道的精确重构。

At present, real and reliable underground coal mine geographic information is an important prerequisite for realizing intelligent mining in coal mines, and real-time roadway data refreshing and reconstruction is an important component of upgrading underground geographic information, however, by the narrow space of coal mine roadways, low and uneven illumination, visual limitations, and the dynamic mining accompanied by the influence of complex environments such as dust, water mist, etc., which leads to poor visual and LiDAR and other perceptions of poor accuracy, low reliability, and adaptive capacity Insufficient, it is difficult to achieve high-precision roadway real-time characterization. Millimeter-wave radar has high adaptability to dust, water mist and other complex environments. Based on this, we propose the research of digital modelling of coal mine roadway by millimeter-wave radar array, which can carry out three-dimensional scanning and measurement and accurate reconstruction of coal mine roadway, provide accurate spatial information of the roadway and geological features in real time, and provide important support for intelligent mining in coal mines, and accurate positioning and orientation of equipment, as well as path planning.

Aiming at the problem of complex environment and difficult information acquisition in coal mine roadway, we study the influence law of environmental factors such as dust, water mist, and roadway structure on millimetre wave radar signals, and build a simulated environment experiment platform. Collect the millimeter wave radar signal information data in different (state, particle size and concentration) dust environments and different water mist environments, establish the millimeter wave signal attenuation model of coal mine roadway enclosure, and compare and analyse the perception mechanism of millimeter wave radar in three environmental factors. Research and analysis of different states of dust particle size and concentration, different shapes of water mist particle size and spraying area and the ideal roadway structure of the millimeter wave radar sensing characteristics, the results show that the different static dust particle size and concentration of the millimeter wave radar signal are not obvious, but the dynamic dust concentration, the larger the particle size, the larger the millimeter wave radar signal, the more obvious the impact of the millimeter wave radar signal; in the state of the water mist, the particle size of the water mist and spraying area of the millimeter wave radar signal can be less impact. Millimetre wave radar signal impact is small can not be considered, but the water droplets, water column on the millimetre wave radar signal impact is more obvious; ideal roadway rock environment on the millimetre wave signal attenuation is small, relative to the radar itself signal loss is negligible. It can be seen that the millimeter wave radar has high adaptability under the complex environmental conditions of coal mine roadway, providing high-precision coal mine roadway point cloud data for the construction of three-dimensional roadway spatial model of coal mine, and providing a new method for digital modelling of roadway.

Aiming at the roadway digital modelling problem of millimeter-wave radar array under the complex environment of coal mine, firstly, we analyse the characteristics of single-frame data of millimeter-wave point cloud, and put forward the millimeter-wave radar point cloud multi-layer filtering and noise reduction processing and dynamic sub-map alignment method, filtering out the outliers, isolated points and false target points, and realizing the single-radar point cloud alignment; secondly, in order to solve the problem of single millimeter-wave radar not being able to get the complete roadway's enclosing rock information at one time. Secondly, in order to solve the problem that a single millimetre-wave radar cannot obtain the complete roadway enclosure rock information at one time, we combine the millimetre-wave radar elevation field of view characteristics, construct the millimetre-wave radar array sensing method with improved Iterative Closest Point (ICP) alignment, and put forward the fusion model of the roadway point cloud of millimetre-wave radar arrays. The roadway surrounding rock contour, close to the actual roadway shape and size, to achieve the accurate reconstruction of the digital roadway model. Through experimental analysis, the average point cloud density of the roadway is improved by 29.1% using the point cloud density weighted method, which solves the unevenness of the point cloud data of the millimetre wave radar array during the reconstruction of the roadway model and improves the average density of the point cloud.

Aiming at the problems of spatial inhomogeneity of millimeter-wave radar array point cloud and incomplete reconstruction model, we analyse the principles of attention mechanism and generative adversarial network and their advantages, and put forward the optimization method of digital modeling of coal mine roadway based on attention mechanism and generative adversarial network. A dual-mode hierarchical sampling unit and a differential feature expansion unit are designed to capture point cloud features at different levels and achieve point cloud feature aggregation. Considering the density and spatial distribution of millimetre wave point cloud, a spatial attention mechanism is proposed to correlate the density and spatial location features, and pay attention to the potential local point cloud features. A dual discriminator model based on the spatial attention mechanism is established, and the training pays attention to the local point cloud features and global spatial information to generate high-quality roadway point cloud data and improve the accuracy of the roadway spatial coordinates. Through experimental analysis, the optimization method of coal mine roadway digital modelling based on attention mechanism and generative adversarial network can improve the density and uniformity of point cloud, realize point cloud densification and complementation, and reconstruct the roadway to be more perfect and smooth.

Through laboratory experimental testing and simulated roadway environment verification, based on revealing the perception mechanism of millimeter-wave radar arrays under complex environmental conditions in coal mines, millimeter-wave radar arrays-based digital modeling of coal mine roadways is realized, and the results show that the method can fully demonstrate the real roadway surrounding rock information, with an average absolute error percentage of width of 0.82% and an average absolute error percentage of height of 0.72%. Through the optimization method of coal mine roadway digital modelling based on attention mechanism and generative adversarial network, the average absolute error of width after roadway reconstruction is 2.02cm, which is improved by 28% relative to the pre-optimization period, and the average absolute error of height is 1.52cm, which is improved by 23% relative to the pre-optimization period, which achieves the accurate reconstruction of coal mine roadways.

[1]雷晓荣．“孔-井-地”一体化智能钻进系统及关键技术[J]．煤炭科学技术，2020，48(07)：274-281．

[2]王海军，刘再斌，雷晓荣等．煤矿巷道三维激光扫描关键技术及工程实践[J]．煤田地质与勘探，2022，50(01)：109-117．

[3]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．

[4]李梅，姜展，姜龙飞，孙振明．三维可视化技术在智慧矿山领域的研究进展[J]．煤炭科学技术，2021，49(02)：153-162．

[5]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．

[6]陈先中，刘荣杰，张森等．煤矿地下毫米波雷达点云成像与环境地图导航研究进展[J]．煤炭学报，2020，45(06)：2182-2192．

[7]王存飞，荣耀．透明工作面的概念、架构与关键技术[J]．煤炭科学技术，2019，47(07): 156-163．

[8]菅洁，谢建林，郭勇义．煤矿井下粉尘浓度与粉尘粒度测定分析[J]．太原理工大学学报，2017，48(4)：592−597．

[9]梁运涛，田富超，冯文彬，等．我国煤矿气体检测技术研究进展[J]．煤炭学报，2021，46(06)：1701-1714．

[10]胡兴涛，朱涛，苏继敏，等．煤矿巷道智能化掘进感知关键技术[J]．煤炭学报，2021，46(07)：2123-2135．

[11]Ranjan A, Sahu H B, Misra P. MineSense: sensing the radio signal behavior in metal and non-metal underground mines[J]. Wireless Networks, 2019, 25: 3643-3655.

[12]蒋俊林，陈昶昊，姚志强，李方言，何晓峰，潘献飞．烟雾环境下单通道毫米波雷达建图算法研究[C]//2022中国自动化大会论文集，2022：153-158．

[13]王战古．不良天气条件下车辆检测方法研究[D]．吉林大学，2022．

[14]Zang S, Ding M, Smith D, et al. The impact of adverse weather conditions on autonomous vehicles: How rain, snow, fog, and hail affect the performance of a self-driving car[J]. IEEE vehicular technology magazine, 2019, 14(2): 103-111.

[15]魏力，石丹，崔强，李安平．基于波形设计的车载FMCW雷达干扰抑制方案[C]//第28届全国电磁兼容学术会议论文集，2022：68−73．

[16]Tuley J, Vandapel N, Hebert M. Analysis and removal of artifacts in 3-D LADAR data[C]//Proceedings of the 2005 IEEE International Conference on Robotics and Automation. IEEE, 2005: 2203-2210.

[17]Walther M, Steinhaus P, Dillmann R. A foveal 3D laser scanner integrating texture into range data. [C]// 9th International Conference on Intelligent Autonomous Systems. IAS, 2006: 748-755.

[18]Fekete S, Diederichs M, Lato M. Geotechnical and operational applications for 3-dimensional laser scanning in drill and blast tunnels[J]. Tunnelling and Underground Space Technology, 2010, 25(5): 614-628.

[19]Baker C, Morris A, Ferguson D, et al. A campaign in autonomous mine mapping[C]// IEEE International Conference on Robotics & Automation. IEEE, 2004, 2.

[20]Ferguson D, Morris A, Hahnel D. An Autonomous Robotic System For Mapping Abandoned Mines. Proceedings of The 2003 IEEE International Conference on Robotics and Automation, Taipei, Taiwan, 2003, 9: 4270-4275

[21]Raval S, Banerjee B P, Canbulat I, et al. A Preliminary Investigation of Mobile Mapping Technology for Underground Mining[C]// International Geoscience and Remote Sensing Symposium (IGARSS), 2019: 6071-6074.

[22]姜龙飞，毛善君，李梅，等．基于激光点云的割煤顶板线提取技术研究[J]．煤炭科学技术，2022，50(06)：286-291．

[23]王海军，刘再斌，雷晓荣，韩保山，陆自清．煤矿巷道三维激光扫描关键技术及工程实践[J]．煤田地质与勘探，2022，50(01)：109−117．

[24]蒋盛锋．基于三维激光扫描仪的三维点云地图构建研究[D]．武汉：华中科技大学，2016．

[25]付忠敏．基于激光扫描的井下点云数据采集与预处理系统研究[D]．武汉：华中科技大学，2017．

[26]万松．煤矿井下三维点云边缘检测及配准研究[D]．华中科技大学，2015．

[27]杨健健，王超，张强，等．井工巷道环境建模与掘进障碍检测方法研究[J]．煤炭科学技术，2020，48(11)，12-18．

[28]Di K, Peng M. Wide Baseline Mapping for Mars Rovers[J]. Photogrammetric Engineering & Remote Sensing, 2011, 77(6): 609-618.

[29]王宣银，潘锋，向桂山．基于Snake模型的特定人脸三维重建方法[J]．机械工程学报，2007，43(7)：168−173．

[30]王欣，袁坤，于晓．基于运动恢复的双目视觉三维重建系统设计[J]．光学精密工程，2014(05)：273−281．

[31]Cui H, Xiang G, Shen S, et al. HSfM: Hybrid Structure-from-Motion[C]// IEEE Conference on Computer Vision & Pattern Recognition. IEEE, 2017: 1212-1221.

[32]Huang P H, Matzen K, Kopf J, et al. DeepMVS: Learning Multi-view Stereopsis[J]. IEEE, 2018: 2821-2830.

[33]Yao Y, Luo Z, Li S, et al. Recurrent MVSNet for High-resolution Multi-view Stereo Depth Inference[J]. IEEE, 2019: 5520-5529.

[34]Blaha M, Vogel C, Richard A, et al. Large-scale semantic 3d reconstruction: an adaptive multi-resolution model for multi-class volumetric labeling[C]//Proceedings of the IEEE conference on computer vision and pattern recognition. 2016: 3176-3184.

[35]刘彬，刘学军，张华．线性的增量式三维稀疏重建系统设计[J]．电光与控制，2019，26(7)：100−104，109．

[36]郎雅琨．基于深度学习的图像三维重建的研究[D]．中北大学，2021．

[37]Jakob Engel, Thomas Schöps, Daniel Cremers. LSD-SLAM: Large-Scale Direct Monocular SLAM[C]// European Conference on Computer Vision. Springer, Cham, 2014: 834-849.

[38]Engel Jakob, Koltun Vladlen, Cremers Daniel. Direct Sparse Odometry[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017, 40(3): 611-625.

[39]Henry P, Krainin M, Herbst E, et al. RGB-D mapping: Using Kinect-style depth cameras for dense 3D modeling of indoor environments[J]. International Journal of Robotics Research, 2012, 31(5): 647-663.

[40]Lee D, Kim H, Myung H. GPU-based real-time RGB-D 3D SLAM[C]// International Conference on Ubiquitous Robots & Ambient Intelligence. IEEE, 2012: 46-48.

[41]Schops T, Sattler T, Pollefeys M. Bad slam: Bundle adjusted direct rgb-d slam[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2019: 134-144.

[42]Tan W, Liu H, Dong Z, et al. Robust monocular SLAM in dynamic environments[C]// Mixed and Augmented Reality (ISMAR), 2013 IEEE International Symposium on. IEEE, 2013: 209-218

[43]舒凯翔．基于RGB-D图像的移动机器人三维地图构建与导航系统研究与设计[D]．广州：华南理工大学，2018．

[44]康凯．基于机器视觉的移动机器人定位与三维地图重建方法研究[D]．哈尔滨：哈尔滨工业大学，2017．

[45]Foessel-Bunting A, Bares J, Whittaker W. Three-Dimensional Map Building With MMW Radar[C]// 3rd International Conference on Field and Service Robotics, 2001.

[46]Adams M, Jose E. Millimeter Wave RADAR Power-Range Spectra Interpretation for Multiple Feature Detection[J]. Autonomous Mobile Robots, 2018: 41-98.

[47]Clark S, Durrant-Whyte H. Autonomous land vehicle navigation using millimeter wave radar[C]//Proceedings. 1998 IEEE International Conference on Robotics and Automation (Cat. No. 98CH36146). IEEE, 1998, 4: 3697-3702.

[48]Brooker G, Hennessey R, Bishop M, et al. High-resolution millimeter-wave radar systems for visualization of unstructured outdoor environments[J]. Journal of Field Robotics, 2006, 23(10): 891-912.

[49]Reina G, Milella A, Underwood J. Self-learning classification of radar features for scene understanding[J]. Robotics & Autonomous Systems, 2012, 60(11): 1377-1388.

[50]Rouveure R ,Faure P ,Monod M .Description and experimental results of a panoramic K‐band radar dedicated to perception in mobile robotics applications[J]. Journal of Field Robotics, 2018, 35(5): 678-704.

[51]Reina G, Underwood J, Brooker G, et al. Radar-based perception for autonomous outdoor vehicles[J]. Journal of Field Robotics, 2011, 28(6): 894-913.

[52]Rouveure R, Faure P, Monod M O. Radar imager for perception and mapping in outdoor environments[C]//International Conference on Advanced Concepts for Intelligent Vision Systems. Berlin, Heidelberg: Springer Berlin Heidelberg, 2009: 618-628.

[53]Marck, J.W, Mohamoud, A.A, Houwen, E.H.V.D. Indoor radar SLAM A radar application for vision and GPS denied environments[J]. Euma, 2013, 1552(1): 471-474.

[54]张珂梦．基于毫米波雷达的SLAM关键技术研究[D]．重庆大学，2022．

[55]崔巍杰．毫米波和激光雷达数据融合的SLAM算法研究[D]．电子科技大学，2019．

[56]徐旺．车载毫米波雷达多目标探测与定位构图[D]．哈尔滨工业大学，2019．

[57]陈文亮，王俊，袁常顺．多毫米波雷达坐标标定方法[C]//第十三届全国DSP应用技术学术会议论文集，2021:28-31．

[58]王立成，孔德明，沈阅，曹帅，张钰．堆取料机防碰检测系统多毫米波雷达标定方法研究[J]．燕山大学学报，2022，46(03):246-256．

[59]赖欣欣．毫米波交通雷达多目标跟踪算法研究与应用[D]．厦门大学，2018．

[60]Zhang R, Cao S. Real-Time Human Motion Behavior Detection via CNN Using mmWave Radar[J]. IEEE Sensors Letters, 2018: 1-4.

[61]鞠夕强，孟文，孟祥印．一种改进的毫米波雷达聚类算法[J]．科学技术与工程，2021，21(20)：8537-8543．

[62]宋佳．多毫米波雷达数据关联及融合算法研究[D]．吉林大学，2022．

[63]QI C R, Su H, Mo K, et al. Pointnet: deep learning on point sets for 3d classification and segmentation[C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2017: 652-660.

[64]Fan H, Su H, Guibas L J. A point set generation network for 3d object reconstruction from a single image[C]//Proceedings of the IEEE conference on computer vision and pattern recognition. 2017: 605-613.

[65]Lin C H, Kong C, Lucey S. Learning efficient point cloud generation for dense 3d object reconstruction[C]//proceedings of the AAAI Conference on Artificial Intelligence. 2018, 32(1).

[66]Groueix T, Fisher M, Kim V G, et al. A papier-mâché approach to learning 3d surface generation[C]//Proceedings of the IEEE conference on computer vision and pattern recognition. 2018: 216-224.

[67]Qi C R, Yi L, Su H, et al. Pointnet++: Deep hierarchical feature learning on point sets in a metric space[J]. Advances in neural information processing systems, 2017, 30.

[68]YU L, LI X, FU C W, et al. Pu-net: point cloud upsampling network[C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2018: 2790-2799.

[69]Corsini M, Cignoni P, Scopigno R. Efficient and flexible sampling with blue noise properties of triangular meshes[J]. IEEE transactions on visualization and computer graphics, 2012, 18(6): 914-924.

[70]Zhang W, Jiang H, Yang Z, et al. Data-driven upsampling of point clouds[J]. Computer-Aided Design, 2019, 112: 1-13.

[71]Yu L, Li X, Fu C W, et al. Ec-net: an edge-aware point set consolidation network[C]//Proceedings of the European conference on computer vision (ECCV). 2018: 386-402.

[72]Kim J, Lee J K, Lee K M. Accurate image super-resolution using very deep convolutional networks[C]//Proceedings of the IEEE conference on computer vision and pattern recognition. 2016: 1646-1654.

[73]Yifan W, Wu S, Huang H, et al. Patch-based progressive 3d point set upsampling[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2019: 5958-5967.

[74]Wang W, Huang Q, You S, et al. Shape inpainting using 3d generative adversarial network and recurrent convolutional networks[C]//Proceedings of the IEEE international conference on computer vision. 2017: 2298-2306.

[75]Li R, Li X, Fu C W, et al. Pu-gan: a point cloud upsampling adversarial network[C]//Proceedings of the IEEE/CVF international conference on computer vision. 2019: 7203-7212.

[76]Qian Y, Hou J, Kwong S, et al. PUGeo-Net: A geometry-centric network for 3D point cloud upsampling[C]//European conference on computer vision. Cham: Springer International Publishing, 2020: 752-769.

[77]Wen X, Xiang P, Han Z, et al. Pmp-net++: Point cloud completion by transformer-enhanced multi-step point moving paths[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2022, 45(1): 852-867.

[78]Simonovsky M, Komodakis N. Dynamic edge-conditioned filters in convolutional neural networks on graphs[C]//Proceedings of the IEEE conference on computer vision and pattern recognition. 2017: 3693-3702.

[79]Qian G, Abualshour A, Li G, et al. Pu-gcn: Point cloud upsampling using graph convolutional networks[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2021: 11683-11692.

[80]姚善化．复杂矿井巷道中电磁波传播特性及相关技术研究[D]．合肥：安徽大学，2010．

[81]易见兵，彭鑫，曹锋等．多尺度特征融合的点云配准算法研究[J/OL]．广西师范大学学报(自然科学版)，1-12[2024-01-28]．

[82]王玉亮，刘飞，沈建新等．基于单目视觉的视网膜三维重建技术研究[J]．中国机械工程，2016，27(18)：2477-2481．