煤矿巷道围岩隐蔽病害随着时间推移缓慢发育，会严重损害巷道围岩结构和支护的稳定性，从而导致顶板事故和水灾事故的发生。因此，提前识别巷道围岩中潜在的灾害隐患对矿山安全生产具有重要意义。本文使用探地雷达这种无损检测方式探测巷道围岩病害，并研究基于深度学习的探地雷达图像智能解译方法，以实现巷道围岩隐蔽病害的智能化检测，最后利用屯兰矿获取的实测数据对本文方案有效性进行了验证。主要研究内容及结论如下：

（1）提出了一种基于GprMax的非均匀介质正演模拟方法。为精确模拟素喷混凝土支护和锚喷网联合支护条件下的支护层脱空、围岩裂隙和围岩含水层病害，本文在简单正演模拟方法的基础上，按照围岩介质的实际组成比例提出了非均匀介质正演模拟方法。通过对不同支护条件下各类病害的雷达图像特征进行细致分析，揭示了病害的内在规律与差异。实验结果表明：本文所提方法在背景噪声，图像特征表达等方面优于传统的正演模拟方法，模拟结果更接近实测雷达数据。这解决了传统的正演模拟获得的探地雷达图像与实际偏差较大的问题，总结出的解译准则对巷道围岩病害的智能化解译提供了理论基础。

（2）探究了DCGAN用于巷道围岩隐蔽病害探地雷达图像数据增强算法的可行性。针对深度学习在巷道围岩病害识别中面临的探地雷达数据样本稀缺和传统数据增强方法导致的样本丰富度不足问题，本文基于DCGAN网络进行改进，结合谱范数归一化和自注意力机制优化原始网络，改善了DCGAN容易出现模式崩溃和训练不稳定的缺陷，实现了对有限的正演数据和少量实测数据进行数据增强的目的。实验结果表明：经过改进的网络能够生成形态丰富、病害特征明显的样本数据，且未对探地雷达图像的原始数据形态造成破坏。在FID、SSIM和LPIPS三个评价指标方面，网络性能提升显著，在探地雷达图像数据增强方面的表现出优异效果。结合所有可用的数据，本文成功构建了包含13074个病害样本的GPR病害数据集，做为巷道围岩病害智能解译的数据支撑。

（3）构建了巷道围岩隐蔽病害探地雷达图像智能检测模型—YOLOv7-GPR。针对现有深度学习模型在巷道围岩病害识别中存在的精度低和定位不准确问题，本文在深入研究YOLOv7模型的基础之上，通过引入全局注意力机制，显著提升了模型的特征提取能力和计算效率；将坐标注意力机制与MP模块相结合，有效提高了模型对目标位置和特征的捕捉精度；采用可变形卷积网络替换ELAN模块中的卷积层，进一步增强了模型对局部特征的捕捉能力；为解决原模型中损失函数优化方式不合理的问题，本文利用EIOU进行改进，优化了损失函数的计算方式。通过消融实验和对比实验证明，相较于原模型，YOLOv7-GPR在mAP指标方面提升了8.3%；此外，YOLOv7-GPR的检测精度明显高于其他经典的目标检测方法，证明了本文方法的优越性。在屯兰矿的巷道实测数据验证中，该方法准确识别了巷道围岩病害，为矿井巷道质量维护提供了有力支持。本文所提方法不仅具有理论价值，更在实际工程应用中展现出较高的实用性。

The concealed diseases in the surrounding rock of coal mine roadways gradually develop over time, which seriously damage the stability of the roadway surrounding rock structure and support, thus leading to accidents such as roof and flooding. Therefore, it is of great significance to identify potential disaster hazards in the roadway rock in advance for the safe production of mines. In this paper, we use ground penetrating radar, a non-destructive testing method, to detect roadway perimeter rock diseases, and study the intelligent interpretation method of ground-penetrating radar images based on deep learning to realize the intelligent detection of hidden diseases in roadway perimeter rock, and verify the validity of this paper's scheme by using the measured data obtained from Tunlan Mine. The main research contents and conclusions are as follows:

(1) An orthotropic simulation method based on GprMax for non-uniform media is proposed. In order to accurately simulate the support layer dehollowing, perimeter rock fissures and perimeter rock aquifer diseases under the conditions of plain spray concrete support and anchor spray network joint support, this paper proposes a non-uniform medium orthogonal simulation method according to the actual composition ratio of the perimeter rock medium on the basis of simple orthogonal simulation method. By carefully analyzing the radar image characteristics of various types of diseases under different support conditions, the intrinsic laws and differences of the diseases are revealed. The experimental results show that the proposed method is better than the traditional forward simulation method in terms of background noise and image features, and the simulation results are closer to the measured radar data. This solves the problem that the ground penetrating radar image obtained by simple forward simulation has a large deviation from the actual one, and the summarized interpretation guideline provides a theoretical basis for the intelligent interpretation of roadway rock damage.

(2) The feasibility of DCGAN for ground-penetrating radar image data enhancement algorithms for hidden diseases in roadway enclosure rocks is explored. In response to the scarcity of ground penetrating radar data samples and the insufficient sample richness caused by traditional data augmentation methods in deep learning for identifying tunnel surrounding rock diseases, this paper improves the DCGAN network by combining spectral norm normalization and self attention mechanism to optimize the original network. This improves the shortcomings of DCGAN, which are prone to pattern collapse and training instability, and achieves the goal of data augmentation on limited forward data and a small amount of measured data. The experimental results show that the improved network is able to generate sample data with rich morphology and obvious disease characteristics without damaging the original data morphology of ground penetrating radar images. In the three evaluation indexes of FID, SSIM and LPIPS, the performance of the network is significantly improved, showing its excellent effect in ground-penetrating radar image data enhancement. Combining all the available data, this paper successfully constructs a GPR disease dataset containing 13074 disease samples, which is used as the data support for the intelligent interpretation of roadway perimeter rock disease.

(3) The YOLOv7-GPR, an intelligent detection model of ground penetrating radar image for hidden diseases in roadway surrounding rock was constructed. Aiming at the low precision and inaccurate localization of the existing deep learning networks in the identification of roadway perimeter rock diseases, based on in-depth research on the YOLOv7 model, this article significantly improves the model's feature extraction ability and computational efficiency by introducing a global attention mechanism; combining the coordinate attention mechanism with the MP module effectively improves the model's capture accuracy of the target location and features; using deformable convolutional network to replace the convolutional layer in the ELAN module further strengthens the model's ability to capture the local features; in order to solve the problem of loss function optimization in the original network, this paper uses the EIOU to improve the optimization of the loss function. The ablation and comparison experiments prove that compared with the original network, YOLOv7-GPR improves 8.3% in mAP; moreover, the detection precision of YOLOv7-GPR is significantly higher than that of other classical target detection methods, which proves the superiority of the method in this paper. In the validation of the measured data of the roadway in Tunlan Mine, the method accurately identifies the roadway peripheral rock disease, which provides a strong support for the maintenance of the quality of the roadway in the mine. The method proposed in this paper not only has theoretical value, but also shows high practicability in practical engineering application.

参考文献

[1] 刘浩. 煤矿井下隐患智能识别模型及其集成系统研究[D]. 北京: 中国矿业大学(北京), 2022.

[2] 薛道成. 煤矿巷道锚杆无损检测技术及在西山矿区的应用研究[D]. 徐州: 中国矿业大学, 2013.

[3] 汪亚林. 基于煤矿五大灾害的应急管理能力评价研究[D]. 淮南: 安徽理工大学, 2021.

[4] 刘敦文. 地下岩体工程灾害隐患雷达探测与控制研究[D]. 长沙: 中南大学, 2001.

[5] 徐俊. 地下工程结构多类型缺陷的雷达信号自动辨识方法与工程应用[D]. 北京: 北京科技大学, 2019.

[6] 龚致民. 公路结构病害探地雷达图像智能识别与信号处理研究[D]. 重庆: 重庆大学, 2020.

[7] Young J. A brief history of GPR fundamentals and applications[J]. Proc. int. conf. ground Penetrating Radar, 1996.

[8] 李华, 鲁光银, 何现启, 等. 探地雷达的发展历程及其前景探讨[J]. 地球物理学进展, 2010, 25(04): 1492-1502.

[9] Momayez M, Hassani F P, Hara A, et al. Application of GPR in Canadian mines[J]. Cim(canadian Mining and Metallurgical) Bulletin, 1996, 89(1001): 107-110.

[10] Strange A D, Ralston J C, Chandran V. Application of ground penetrating radar technology for near-surface interface determination in coal mining[J]. IEEE, 2005, 5: v/701-v/704.

[11] Francke J, Utsi V, Singh K K K. The design of an intrinsically safe ground penetrating radar for detecting abandoned workings in underground coal mines[C]//2012 14th International Conference on Ground Penetrating Radar (GPR): Ieee, 2012: 125-130.

[12] Tolooiyan A, Dyson A P, Karami M, et al. Application of Ground Penetrating Radar (GPR) to Detect Joints in Organic Soft Rock[J]. Geotechnical Testing Journal, 2019, (2).

[13] 刘波,吕玉增. 地质雷达在国内的应用现状及发展[J]. 煤炭技术, 2020, 39(06): 60-63.

[14] 宋雷,黄家会,南生辉. 地质雷达用于探测煤田自燃区的研究[J].煤炭科学技术,1999, (12):23-24, 52.

[15] 刘传孝. 探地雷达方法应用于煤矿井下断层探测[J]. 工程地质学报, 2000, 8(3): 361-363.

[16] 李纯洁,孔德森,王立才,等. 探地雷达在松动圈确定与巷道支护参数优化中的应用[J]. 山东科技大学学报(自然科学版), 2008, (01): 19-22.

[17] 程久龙,潘冬明,李伟,等. 强电磁干扰区灾害性采空区探地雷达精细探测研究[J]. 煤炭学报, 2010, 35(2): 227-231.

[18] 崔凡,耿晓航,俞慧婷,等. 基于探地雷达的煤层小构造超前探测[J]. 煤矿安全, 2019, 50(05): 153-157.

[19] 梁庆华,吴燕清,宋劲,等. 探地雷达在煤巷掘进中超前探测试验研究[J]. 煤炭科学技术, 2014, 42(5): 91-94.

[20] 程建远,陆自清,蒋必辞,等. 煤矿巷道快速掘进的“长掘长探”技术[J]. 煤炭学报, 2022, 47(01): 404-412.

[21] 禾永. 煤矿井下探地雷达应用试验研究[J]. 建井技术, 2017, 38(06): 33-35, 43.

[22] 徐光,徐斌,曲延伦,等. 深部大变形巷道应力缓释与围岩加固联合支护技术研究[J]. 煤矿现代化, 2018, (03): 93-96, 100.

[23] 廖旭涛. 基于城市地质灾害勘查的探地雷达成像和电磁谐振识别技术研究[D].合肥: 合肥工业大学, 2019.

[24] Craig, Warren, Antonios, et al. gprMax: Open source software to simulate electromagnetic wave propagation for Ground Penetrating Radar[J]. Computer Physics Communications, 2016, 209: 163-170.

[25] Verma S, Srivastava S, Bhattacharya B B. Ground Penetrating Radar Response Over Coal Seam Discontinuities[C]//Agu Spring Meeting Abstracts, 2007: NS33A-05.

[26] Jianlin X, Jialin X U, Xiaolin L I. Two-dimensional forward numerical simulation of GPR in separation detection of roof rock[J]. Coal Engineering, 2011, 207: 125-127.

[27] Jiang B, Cheng J, Li P, et al. Using BHR to detect coal seam interface in coal mine[J]. Journal of Applied Geophysics, 2023, 209: 104894.

[28] Zhang D, Tang R, Wen H, et al. Propagation of ground penetrating radar waves in Chinese coals[J]. Plos One, 2020, 15(5): e0233434.

[29] 李敏瑞,马锦辉. GPRMAX~(2D)在巷道质量检测中的应用[J]. 煤矿安全, 2013, 44(09): 166-168.

[30] 郑龙金,唐小泉,曹平华,等. 探地雷达对采煤工作面顶底板复杂构造探测研究[J]. 煤炭技术, 2018, 37(07): 124-127.

[31] 占文锋. 基于FDTD的矿井异常体地质雷达图像正演模拟分析[C]//中国煤炭学会矿井地质专业委员会成立三十五周年暨中国煤炭学会矿井地质专业委员会2017年学术论坛, 2017: 278-288.

[32] 文虎,张铎,郑学召,等. 基于FDTD的电磁波在煤中传播特性[J]. 煤炭学报, 2017, 42(11): 2959-2967.

[33] 王树奇,宁箭,王振. 基于FDTD的煤层工作面上保护层检测正演分析[J]. 采矿技术, 2020, 20(01): 145-149.

[34] 王诚龙,刘万里,张学亮,等. 基于探地雷达的工作面异常体精准探测技术[J]. 岩土力学, 2022, 43(11): 3198-3208.

[35] 周辉林,姜玉玲,徐立红,等. 基于SVM的高速公路路基病害自动检测算法[J]. 中国公路学报, 2013, 26(02): 42-47.

[36] Maas C, Schmalzl J. Using pattern recognition to automatically localize reflection hyperbolas in data from ground penetrating radar[J]. Computers & geosciences, 2013, 58: 116-125.

[37] 李博,赵永辉,胡书凡,等. 基于哈希算法的地下管线探地雷达图像智能识别[J]. 地球物理学进展, 2022, 37(01): 386-396.

[38] Besaw L E, Stimac P J. Deep learning algorithms for detecting explosive hazards in ground penetrating radar data[C]//Detection and Sensing of Mines, Explosive Objects, and Obscured Targets XIX. SPIE, 2014, 9072: 224-234.

[39] Dou Q, Wei L, Magee D R, et al. Real-Time Hyperbola Recognition and Fitting in GPR Data[J]. Ieee Transactions on Geoscience and Remote Sensing, 2016, PP(1): 1-12.

[40] Pham M T, Lefèvre S. Buried object detection from B-scan ground penetrating radar data using Faster-RCNN[C]//IGARSS 2018-2018 IEEE international geoscience and remote sensing symposium. IEEE, 2018: 6804-6807.

[41] Zhou X, Chen H, Li J. An automatic GPR B-scan image interpreting model[J]. IEEE Transactions on Geoscience and Remote Sensing, 2018, 56(6): 3398-3412.

[42] Qin H, Zhang D, Tang Y, et al. Automatic recognition of tunnel lining elements from GPR images using deep convolutional networks with data augmentation[J]. Automation in Construction, 2021, 130: 103830.

[43] Xiong X, Meng A, Lu J, et al. Automatic detection and location of pavement internal distresses from ground penetrating radar images based on deep learning[J]. Construction and Building Materials, 2024, 411: 134483.

[44] Li S, Gu X, Xu X, et al. Detection of concealed cracks from ground penetrating radar images based on deep learning algorithm[J]. Construction and Building Materials, 2021, 273: 121949.

[45] Gao R, Zhu H, Liao Q, et al. Detection of coal fire by deep learning using ground penetrating radar[J]. Measurement, 2022, 201: 111585.

[46] 李宝奇,贺昱曜,郭元术,等. 基于DBN的隧道地质超前预报自动解释算法[J]. 长安大学学报(自然科学版), 2017, 37(03): 90-96.

[47] 冯德山,杨子龙. 基于深度学习的隧道衬砌结构物探地雷达图像自动识别[J]. 地球物理学进展, 2020, 35(04): 1552-1556.

[48] 杨必胜,宗泽亮,陈驰,等. 车载探地雷达地下目标实时探测法[J]. 测绘学报, 2020, 49(07): 874-882.

[49] 王辉,欧阳缮,廖可非,等. 基于深度学习的GPR B-SCAN图像双曲线检测方法[J]. 电子学报, 2021, 49(05): 953-963.

[50] 罗婷倚,杨哲,张军,等. 基于GPR信号和卷积核的水泥路面脱空识别算法[J]. 地球物理学进展, 2022, 37(06): 2580-2588.

[51] 王大为,吕浩天,汤伏蛟,等. 三维探地雷达道路隐性病害检测分析与数字化技术综述[J]. 中国公路学报, 2023, 36(03): 1-19.

[52] 曾昭发. 探地雷达方法原理及应用[M]: 探地雷达方法原理及应用, 2006.

[53] 李夏利. 基于深度学习的隧道衬砌病害识别和分类研究[D]. 南京: 南京邮电大学, 2021.

[54] 李聪聪. 隧道主要不良地质体探地雷达图像智能识别方法研究[D]. 济南: 山东大学, 2022.

[55] Kurtz J L, Fisher III J W, Skau G, et al. Advances in ground-penetrating radar for road subsurface measurements[C]//Radar Sensor Technology II. SPIE, 1997, 3066: 11-21.

[56] Yee K. Numerical solution of initial boundary value problems involving Maxwell's equations in isotropic media[J]. IEEE Transactions on antennas and propagation, 1966, 14(3): 302-307.

[57] Giannopoulos A. Modelling ground penetrating radar by GprMax[J]. Construction and building materials, 2005, 19(10): 755-762.

[58] 张熙俊. 基于探地雷达图像的城市道路地下病害识别方法研究[D]. 西安: 西安电子科技大学, 2022.

[59] 范黎. 基于生成对抗网络的图像数据增强技术研究及应用[D]. 杭州: 浙江大学, 2022.

[60] Radford A, Metz L, Chintala S. Unsupervised representation learning with deep convolutional generative adversarial networks[J]. arXiv preprint arXiv:1511.06434, 2015.

[61] 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.

[62] Redmon J, Farhadi A. YOLO9000: better, faster, stronger[C]//Proceedings of the IEEE conference on computer vision and pattern recognition. 2017: 7263-7271.

[63] Redmon J, Farhadi A. Yolov3: An incremental improvement[J]. arXiv preprint arXiv:1804.02767, 2018.

[64] Bochkovskiy A, Wang C Y, Liao H Y M. Yolov4: Optimal speed and accuracy of object detection[J]. arXiv preprint arXiv:2004.10934, 2020.

[65] 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.

[66] 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.

[67] Wang C Y, Bochkovskiy A, Liao H Y M. YOLOv7: Trainable bag-of-freebies sets new state-of-the-art for real-time object detectors[C]//Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 2023: 7464-7475.

[68] Liu W, Anguelov D, Erhan D, et al. Ssd: Single shot multibox detector[C]//Computer Vision–ECCV 2016: 14th European Conference, Amsterdam, The Netherlands, October 11–14, 2016, Proceedings, Part I 14. Springer International Publishing, 2016: 21-37.

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

[70] 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.

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

[72] Ren S, He K, Girshick R, et al. Faster r-cnn: Towards real-time object detection with region proposal networks[J]. Advances in neural information processing systems, 2015, 28.

[73] Cai Z, Vasconcelos N. Cascade r-cnn: Delving into high quality object detection[C]//Proceedings of the IEEE conference on computer vision and pattern recognition. 2018: 6154-6162.

[74] Cui S, Jiang Y. Effective Lipschitz constraint enforcement for Wasserstein GAN training[C]//2017 2nd IEEE International Conference on Computational Intelligence and Applications (ICCIA). IEEE, 2017: 74-78.

[75] Bahdanau D, Cho K, Bengio Y. Neural machine translation by jointly learning to align and translate[J]. arXiv preprint arXiv:1409.0473, 2014.

[76] Vaswani A, Shazeer N, Parmar N, et al. Attention is all you need[J]. Advances in neural information processing systems, 2017, 30.

[77] D. C. Dowson, B. V. Landau. The Fréchet Distance between Multivariate Normal Distributions[J]. Journal of Multivariate Analysis, 1982, 12(3): 450-455.

[78] Malpica, Bovik. Range image quality assessment by Structural Similarity[C]//Ieee International Conference on Acoustics, 2009: 1149-1152.

[79] Zhang R, Isola P, Efros A A, et al. The unreasonable effectiveness of deep features as a perceptual metric[C]//Proceedings of the IEEE conference on computer vision and pattern recognition. 2018: 586-595.

[80] Liu Y, Shao Z, Hoffmann N. Global attention mechanism: Retain information to enhance channel-spatial interactions[J]. arXiv preprint arXiv:2112.05561, 2021.

[81] Hou Q, Zhou D, Feng J. Coordinate attention for efficient mobile network design[C]//Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 2021: 13713-13722.

[82] Dai J, Qi H, Xiong Y, et al. Deformable convolutional networks[C]//Proceedings of the IEEE international conference on computer vision. 2017: 764-773.

[83] Zhu X, Hu H, Lin S, et al. Deformable convnets v2: More deformable, better results[C]//Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 2019: 9308-9316.

[84] Zhang Y F, Ren W, Zhang Z, et al. Focal and efficient IOU loss for accurate bounding box regression[J]. Neurocomputing, 2022, 506: 146-157.