机械设备在运转过程中往往会出现由于超负荷运行或长期缺乏有效维护造成因故停机的情况，从而给企业带来巨大经济损失和安全隐患。齿轮箱作为机械传动的核心部件迫切需要对其运行过程进行监测与故障诊断，防患于未然。大多数状态监测和故障诊断技术均以设备运行过程中的振动信号为手段，通过分析不同状态下信号频率特征来确定设备状态及故障类型。然而，对设备磨损故障，直接研究磨损产物——磨粒，较振动信号能更直观准确地表达设备磨损状态。磨粒分析一般分为基于典型铁谱图像特征的磨粒识别和基于颗粒浓度的设备剩余寿命预测。本文针对磨粒铁谱图像分类和异常检测展开深入研究，采用卷积神经网络构建基于单一磨粒的铁谱图像分类模型和基于多磨粒的铁谱图像目标检测模型，实现磨粒铁谱图像的智能分类和异常磨粒智能检测。

       首先，使用在线和离线两种方式从待测油液中获取磨粒铁谱图像，根据磨粒分类和异常磨粒检测两种不同情况选择合适的铁谱图像数据集并进行图像标注。本文所得磨粒按形貌特征分为链状磨粒、切削磨粒、疲劳磨粒、球状磨粒和严重滑动磨粒。用于分类问题的数据集包含除球状磨粒之外的其他四种类型，用于异常检测问题的磨粒不包含正常磨损产生的链状磨粒。并且，在异常检测问题中考虑对润滑油中油泥的检测。为解决图像样本数量不足问题，自主设计了基于相似性的虚拟磨粒铁谱图像，并针对不同问题分别构成相应数据集，为深度学习研究奠定基础。

       其次，研究了基于卷积神经网络的最优磨粒铁谱图像分类模型。提出一种虚拟图像作为源数据的迁移学习方法，以解决磨粒图像样本数量较少的问题。以AI Studio为开发平台，使用PaddlePaddle深度学习框架，构建了基于AlexNet的卷积神经网络基础模型，使用参数迁移方法研究不同参数对模型分类准确率的影响，在合理调控范围内寻求使模型取得最佳分类效果的最优参数组合，在测试集上分类准确率达到93.8%。同时，对各卷积层输出的特征图进行可视化，直观分析模型训练过程中卷积神经网络特征提取过程，使用聚类算法表征了磨粒分类结果。

       最后，针对真实工况下的矿用齿轮箱磨粒铁谱图像，提出基于两级迁移学习的异常磨粒智能检测模型。在检测过程中，考虑油泥作为干扰源，以YOLOv3单阶段目标检测算法为基础，使用自主设计的混合磨粒铁谱图像数据集构建基础模型；然后，从迁移学习源数据出发，使用不同虚拟异常磨粒数据集进行比较研究，选择更合适的数据进行迁移学习；最后，使用矿用齿轮箱磨粒铁谱图像，验证上述研究的实用性。使用两级迁移学习优化模型检测效果，分析了优化前后模型误差来源，证明了结果的有效性。经过两级迁移学习，模型在验证集上平均检测精度为86.1%，平均召回率为95.8%。

       总体上，本文以磨粒铁谱图像为研究对象，以深度迁移学习为研究方法，针对矿用齿轮箱磨粒分类与异常检测问题，采用虚拟图像数据作为源数据进行研究，构建了基于虚拟图像与深度迁移学习的磨粒铁谱图像智能分类模型与基于两级迁移学习的异常磨粒检测模型，按照“虚拟数据→公开数据→实测数据”两级迁移学习实现了矿用齿轮箱异常磨粒智能检测，检测精度相比基础模型提高了44.5%。

In the process of operation, mechanical equipment will often be shut down due to overload operation or long-term lack of effective maintenance, which will bring huge economic losses and security risks to enterprises. As the core component of mechanical transmission, gearbox is in urgent need of monitoring and fault diagnosis in its operation process. Most of the condition monitoring and fault diagnosis technologies are based on the vibration signals in the process of equipment operation, and determine the equipment status and fault types by analyzing the signal frequency characteristics under different states. However, for the equipment wear fault, it is more intuitive and accurate to study the wear product abrasive directly than the vibration signal. Wear debris analysis is generally divided into wear debris recognition based on typical ferrographic image features and residual life prediction based on debris concentration. In this paper, the classification and anomaly detection of wear debris ferrographic image are deeply studied. Convolution neural network is used to build the classification model of ferrographic image based on single wear debris and the target detection model of ferrographic image based on multi wear debris, so as to realize the intelligent classification of wear debris ferrographic image and the intelligent detection of abnormal wear debris.

Firstly, the ferrographic images of the wear debris are obtained from the oil to be measured by online and offline methods. According to the classification of wear debris and the detection of abnormal wear debris, the appropriate image data set of ferrography is selected and the image is labeled. The wear debris obtained in this paper are divided into chain wear debris, cutting abrasive debris, fatigue wear debris, spherical wear debris and serious sliding wear debris according to the morphology characteristics.The data set used to classify the problem contains four types of wear debris except spherical ones, and the debris used to detect abnormal problems do not contain chain abrasive particles produced by normal wear. In addition, the detection of mud in lubricating oil is considered in the detection problem. In order to solve the problem of insufficient image sample, a virtual abrasive iron spectrum image based on similarity was designed, and corresponding data sets were constructed for different problems, which laid the foundation for deep learning.

Secondly, the optimal classification model of wear debris ferrographic image based on convolution neural network is studied. In order to solve the problem of small number of wear debris image samples, a transfer learning method based on virtual image as source data is proposed. With AI Studio as the development platform, using PaddlePaddle deep learning framework, the basic model of convolution neural network based on AlexNet is constructed. The influence of different parameters on the classification accuracy of the model is studied by using parameter migration method. The optimal parameter combination to achieve the best classification effect is sought within the reasonable control range. The accuracy rate of classification was 93.8%. At the same time, the output feature map of each convolution layer is visualized to intuitively analyze the feature extraction process of convolution neural network in the process of model training, and the results of wear debris classification are characterized by clustering algorithm.

Finally, aiming at the ferrographic image of wear debris in the mine gearbox under real working conditions, an intelligent detection model of abnormal wear debris based on two-level transfer learning is proposed. In the process of detection, oil sludge is considered as the interference source. Based on the single-stage object detection algorithm of YOLOv3, the basic model is constructed by using the self-designed ferrographic image data set of mixed wear debris. Then, starting from the transfer learning source data, different virtual abnormal wear debris data sets are used for comparative study, and more appropriate data is selected for migration learning. Finally, the practicability of the above research is verified by using the ferrographic image of the wear debris in the mine gearbox. Two level transfer learning is used to optimize the model, and the error sources before and after optimization are analyzed, which proves the effectiveness of the results. After two-level transfer learning, the average detection accuracy of the model in the verification set is 86.1%, and the average recall rate is 95.8%.

In general, in this paper, the ferrographic image of wear debris is taken as the research object, and the depth transfer learning is taken as the research method. Aiming at the problem of classification and detection of wear debris in mine gearbox, the virtual image data is used as the source data. The intelligent classification model of wear debris ferrographic image based on virtual image and deep transfer learning and the abnormal wear debris detection model based on two-level transfer learning are constructed. According to the two-level transfer learning mode of "virtual data → public data → measured data", the intelligent detection of abnormal wear debris in mine gearbox is realized, and the detection accuracy is improved by 44.5% compared with the basic model.

[1]张生旺. 矿用设备润滑状态监测与故障诊断研究[J]. 煤炭与化工，2014, 37(09): 32-34+38.

[2]彭鹏，陈李果，汪久根，等. 机械磨损的检测技术综述[J]. 润滑与密封，2018, 43(1): 115-124.

[3]樊红卫，张旭辉，曹现刚，等. 智慧矿山背景下我国煤矿机械故障诊断研究现状与展望[J]. 振动与冲击，2020, 39(24): 194-204.

[4]王国法，庞义辉，任怀伟. 煤矿智能化开采模式与技术路径[J]. 采矿与岩层控制工程学报，2020， 2(1): 6-20.

[5]王国法，刘峰，孟祥军，等. 煤矿智能化(初级阶段)研究与实践[J]. 煤炭科学技术，2019, 47(8): 1-36．

[6]温诗铸，黄平. 摩擦学原理[M]. 北京：清华大学出版社，2002.

[7]Sun J, Wood R J K, Wang L, et al. Wear monitoring of bearing steel using electrostatic and acoustic emission techniques[J]. Wear, 2005, 259(7-12): 1482-1489.

[8]钟秉林，黄仁. 机械故障诊断学[M]. 北京：机械工业出版社，2007.

[9]Samanta B, Al-Balushi K R. Artificial neural network based fault diagnostics of rolling element bearings using time-domain features[J]. Mechanical Systems & Signal Pocessing, 2003, 17(2): 317-328.

[10]Ali J B, Fnaiech N, Saidi L, et al. Application of empirical mode decomposition and artificial neural network for automatic bearing fault diagnosis based on vibration signals[J]. Applied Acoustics, 2015, 89(3): 16-27.

[11]Saravanan N, Cholairajan S, Ramachan-Dran K I. Vibration-based fault diagnosis of spur bevel gear box using fuzzy technique[J]. Expert Systems with Applications, 2009, 36(2): 3119-3135.

[12]Huang W, Kong F, Zhao X. Spur bevel gearbox fault diagnosis using wavelet packet transform and rough set theory[J]. Journal of Intelligent Manufacturing, 2018, 29: 1257-1271.

[13]Zhang X, Liang Y, Zhou J, et al. A novel bearing fault diagnosis model integrated permutation entropy, ensemble empirical mode decomposition and optimized SVM[J]. Measurement, 2015, 69: 164-179.

[14]Toutountzakis T, Tan C K, Mba D. Application of acoustic emission to seeded gear fault detection[J]. NDT ＆ E International, 2005, 38(1): 27-36.

[15]严新平，谢友柏，萧汉梁. 油液监测技术的研究现状与发展方向[J]. 中国机械工程，1997, 8(1): 102-105.

[16]Peng Y P, Wu T H, Wang S, et al. Wear state identification using dynamic features of wear debris for on-line purpose[J]. Wear, 2017, 376-377: 1885-1891.

[17]Fan B, Li B, Feng S, et al. Modeling and experimental investigations on the relationship between wear debris concentration and wear rate in lubrication systems[J]. Tribology International, 2017, 109: 114-123.

[18]Cao W, Dong G, Xie Y B, et al. Prediction of wear trend of engines via on-line wear debris monitoring[J]. Tribology International, 2018, 120: 510-519.

[19]Feng S, Fan B, Mao J, et al. Prediction on wear of a spur gearbox by on-line wear debris concentration monitoring[J]. Wear, 2015, 336-337: 1-8.

[20]王汉功，陈桂明. 铁谱图像分析理论与技术[M]. 北京：科学出版社，2005.

[21]北京格普检测有限公司.OVF-R1型在线可视铁谱检测仪[EB/OL]. http://www.bjgpjc.com/A/?L-0466045275.Html.

[22]Seifert W W, Westcott V C. A method for the study of wear particles in lubricating oil[J]. Wear, 1972, 21(1): 27-42.

[23]Liu Y, Xie Y B , Yuan C J , et al. Research on an on-line ferrograph[J]. Wear, 1992, 153(2): 323-330.

[24]杨明忠，樊建春，朱涵斌，等. 双功能旋转式铁谱仪的研制与应用[C]. 全国摩擦学术会议，1992: 83-85.

[25]赵方. 油液分析中的在线铁谱技术及信息融合研究[D]. 西安：西安交通大学，1997.

[26]李祖元. 在线式铁谱仪及其应用监测模型研究[D]. 西安：西安交通大学，1990.

[27]刘岩. 在线铁谱技术的理论与实践研究[D]. 北京：清华大学，1997.

[28]王具平. 铁谱技术及其在铁路部门的应用[J]. 铁道机车车辆工人，2000(03): 26-29.

[29]Peng Y P, Wu T H, Wang S, et al. Oxidation wear monitoring based on the color extraction of on-line wear debris[J]. Wear, 2015, 332-333: 1151-1157.

[30]Wang J, Wang G, Cheng L. Texture extraction of wear particles based on improved random Hough transform and visual saliency[J]. Engineering Failure Analysis, 2020, 109.

[31]Kumar A, Ghosh S K. Size distribution analysis of wear debris generated in HEMM engine oil for reliability assessment: A statistical Approach[J]. Measurement, 2019, 131: 412-418.

[32]Cho U, Tichy J A. Quantitative correlation of wear debris morphology: Grouping and classifification[J]. Tribology International, 2000, 33(7): 461-467.

[33]Wang J, Zhang L, Lu F, et al. The segmentation of wear particles in ferrograph images based on an improved ant colony algorithm[J]. Wear, 2014, 311(1-2): 123-129.

[34]Wu T H, Wu H K, Du Y, et al. Imaged wear debris separation for on-line monitoring using gray level and integrated morphological features[J]. Wear, 2014, 316(1-2): 19-29.

[35]Wang J, Wang X. A wear particle identification method by combining Principal component analysis and grey relational analysis[J]. Wear, 2013, 304(1-2): 96-102.

[36]王联君. 基于深度学习的铁谱图像分类[D]. 南京：南京航空航天大学，2018.

[37]Wang J Q, Bi J, Wang L J, et al. A non-reference evaluation method for edge detection of wear particles in ferrograph images[J]. Mechanical Systems and Signal Processing, 2018, 100: 863-876.

[38]吴振锋. 基于磨粒分析和信息融合的发动机磨损故障诊断技术研究[D]. 南京：南京航空航天大学，2001.

[39]王科俊，赵彦东，邢向磊. 深度学习在无人驾驶汽车领域应用的研究进展[J]. 智能系统学报，2018, 13(01): 55-69.

[40]曾祥云. 人工智能推动无人零售崛起[J]. 企业管理，2017(11): 96-98.

[41]任浩，屈剑锋，柴毅，等. 深度学习在故障诊断领域中的研究现状与挑战[J]. 控制与决策，2017, 32(8): 1345-1358.

[42]张巧丽，赵地，迟学斌. 基于深度学习的医学影像诊断综述[J]. 计算机科学，2017, 44(S2): 11-17.

[43]刘国光. 基于改进支持向量机的磨粒铁谱识别的研究[J]. 润滑与密封，2005(3): 94-96.

[44]周伟，景博，邓森，等. 基于IGA和LS-SVM的航空发动机磨粒识别[J]. 润滑与密封，2013, 38(01): 14-18.

[45]邱丽娟. 基于支持向量机的铁谱磨粒图像识别技术[D]. 太原：太原理工大学，2015.

[46]吕纯，张培林，杨玉栋，等. 一种用于磨粒识别的基于改进PSO算法的支持向量机模型[J]. 润滑与密封，2016, 41(02): 81-85.

[47]Yuan W, Chin K S, Hua M, et al. Shape classification of wear particles by image boundary analysis using machine learning algorithms[J]. Mechanical Systems and Signal Processing, 2016, 72-73: 346-358.

[48]Peng Y P, Wu T H, Cao G Z, et al. A hybrid search-tree discriminant technique for multivariate wear debris classification[J]. Wear, 2017, 392: 152-158.

[49]李加伟，刘晓卫，王崴，等. 基于极限学习机的铁谱磨粒模式识别[J]. 润滑与密封, 2019, 44(06): 72-77.

[50]任松，徐雪茹，欧阳汛，等. 基于分层模糊支持向量机的油液磨粒自动识别[J]. 润滑与密封，2019, 44(05): 1-8.

[51]闫建阳，陈小虎，陈俊康．基于铁谱图像异类特征融合的磨损类型识别方法[J]. 润滑与密封，2020, 45(3): 113-20．

[52]梁华，席忠民，杨明忠. 基于图像处理与神经网络的磨粒识别方法研究[J]. 武汉理工大学学报(信息与管理工程版)，1995(02): 19-24.

[53]Xu K, Luxmoore A R, Deeravi F. Comparison of shape features for the classification of wear particles[J]. Engineering Applications of Artificial Intelligence, 1997, 10(5): 485-493.

[54]刘粲，谢小鹏，陆丕清，等. 基于BP神经网络的铁谱磨粒图像识别方法研究[J]. 润滑与密封，2010, 35(04): 72-75.

[55]林景栋，吴欣怡，柴毅，等. 卷积神经网络结构优化综述[J]. 自动化学报，2020, 46(01): 24-37.

[56]Wu H K, Ngai M K, Liu S L, et al. Restoration of defocused ferrograph images using a large kernel convolutional neural network[J]. Wear, 2019, 426: 1740-1747.

[57]Szátmari I, Schultz A, Rekeczky C, et al. Morphology and autowave metric on CNN Applied to bubble-debris classification[J]. IEEE Transactions on Neural Networks, 2000, 11(6): 1385-1393.

[58]Peng P, Wang J. Wear particle classification considering particle overlapping[J]. Wear, 2019, 422: 119-127.

[59]Peng P, Wang J. FECNN: A Promising model for wear particle recognition[J]. Wear, 2019, 432.

[60]Wang S, Wu T H, Shao T, et al. Integrated model of BP neural network and CNN algorithm for automatic wear debris classification[J]. Wear, 2019, 426: 1761-1770.

[61]张光远. 基于综合特征提取和深度学习的风电机组齿轮箱磨粒分类研究[D]. 秦皇岛：燕山大学，2020.

[62]Wang J Q, Liu X, Wu M, et al. Direct detection of wear conditions by classification of ferrograph images[J]. Braz. Soc. Mech. Sci. Eng. 2020, 42(4).

[63]安超，魏海军，刘竑，等. 基于卷积神经网络的铁谱磨粒智能识别研究[J]. 现代制造工程，2019(07): 111-114.

[64]Peng Y P, Cai J H, Wu T H, et al. A hybrid convolutional neural network for intelligent wear particle classifification[J]. Tribology International, 2019, 138: 166-173.

[65]Peng Y P, Cai J H, Wu T H, et al. Online wear characterisation of rolling element bearing using wear particle morphological features[J]. Wear, 2019, 430: 369-375.

[66]Dollar P, Wojek C, Schiele B, et al. Pedestrian detection: A benchmark[C]. 2009 IEEE Conference on Computer Vision and Pattern Recognition. IEEE, 2009: 304-311.

[67]Uijlings J R, Van D S K E A, Gevers T, et al. Selective search for object recognition[J]. International Journal of Computer Vision, 2013, 104(2): 154-171.

[68]Vedald A, Gulshan V, Varma M, et al. Multiple kernels for object detection[C]. 2009 IEEE 12th International Conference on Computer Vision. IEEE, 2009: 606-613.

[69]Yu Y, Zhang J, Huang Y, et al. Object detection by context and boosted HOG-LBP[C]. ECCV Workshop on PASCAL VOC, 2010.

[70]Zitnick C L, Dollar P. Edge boxes: Locating object Proposals from edges[C]. European Conference on Computer Vision. Springer, Cham, 2014: 391-405.

[71]Hu Q, Zhai L. RGB-D image multi-target detection method based on 3D DSF R-CNN[J]. International Journal of Pattern Recognition and Artificial Intelligence, 2019, 33(8).

[72]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.

[73]Girshick R .Fast R-CNN[C]. 2015 IEEE International Conference on Computer Vision (ICCV). IEEE, 2016: 1440-1448.

[74]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.

[75]Dai J, Li Y, He K, et al. R-fcn: Object detection via region-based fully convolutional networks[C]. Advances in Neural Information Processing Systems. 2016: 379-387.

[76]He K, Gkioxari G, Piotr D, et al. Mask R-CNN[C]. 2017 IEEE International Conference on Computer Vision(ICCV). IEEE, 2017: 2980-2988.

[77]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.

[78]Liu W, Anguelov D, Erhan D, et al. Ssd: Single shot multibox detector[C]. European Conference on Computer Vision. Springer, Cham, 2016: 21-37.

[79]吴鸣. 基于卷积神经网络和图像显著性的铁谱磨粒分析[D]. 南京：南京航空航天大学，2019.

[80]杨智宏，贺石中，冯伟，等. 基于Mask R-CNN网络的磨损颗粒智能识别与应用[J]. 摩擦学学报，2021, 41(01): 105-114.

[81]Peng Y P, Cai J H, Wu T H, et al. WP-DRnet: A novel wear particle detection and recognition network for automatic ferrograph image analysis[J]. Tribology International, 2020, 151.

[82]LeCun Y, Bengio Y, Hinton G. Deep Learning[J]. Nature, 2015, 521(7553): 436-444.

[83]周访滨，邹联华，刘学军，等. 栅格DEM微地形分类的卷积神经网络法[J/OL]. 武汉大学学报(信息科学版): 1-9[2021-03-25]. https://doi.org/10.13203/j.whugis20190311.

[84]刘颖，刘红燕，范九伦. 基于深度学习的小目标检测研究与应用综述[J]. 电子学报，2020, 48(3): 590-601.

[85]Ito S, Chen P, Comte P, et al. Fabrication of screen-printing pastes from TiO2 powders for dye-sensitised solar cells[J]. Progress in Photovoltaics: Research and Applications, 2007, 15(7): 603-612.

[86]Marris H, Deboudt K, Augustin P, et al. Fast changes in chemical composition and size distribution of fine particles during the near-field transport of industrial plumes[J]. Science of the Total Environment, 2012, 427: 126-138.

[87]Lin T Y, Maire M, Belongie S, et al. Microsoft coco: Common objects in context[C]. European Conference on Computer Vision. Springer, Cham, 2014: 740-755.

[88]Deng J, Dong W, Socher R, et al. Imagenet: A large-scale hierarchical image database[C]. 2009 IEEE Conference on Computer Vision and Pattern Recognition. IEEE, 2009: 248-255.

[89]Li Z, Zhou F. FSSD: feature fusion single shot multibox detector[EB/OL]. https://arxiv.org/pdf/ 1712.00960.pdf.

[90]Redmon J, Farhadi A. Yolov3: An incremental improvement[EB/OL]. https://arxiv.org/pdf/1804.02767 v1.pdf.

[91]Redmon J, Farhadi A. YOLO9000: Better, Faster, Stronger[C]. IEEE Conference on Computer Vision & Pattern Recognition. IEEE, 2017: 6517-6525.

[92]Kim K J, Kim P K, Chung Y S, et al. Performance enhancement of YOLOv3 by adding Prediction layers with spatial pyramid pooling for vehicle detection[C]. 2018 15th IEEE International Conference on Advanced Video and Signal Based Surveillance (AVSS). IEEE, 2018: 1-6.

[93]深入浅出Yolov3和Yolov4[EB/OL]. https://mp.weixin.qq.com/s?__biz=MzUyMjE2MTE0Mw%3D% 3D&chksm=f9d14616cea6cf008515d3a99179771f577e1e5e5269dd132751cd8184d6c0adfa93ed2f93ca&idx=1&mid=2247489678&scene=21&sn=c8bd061cea81c57d95a02987073f085b#wechat_ redirect.

[94]吴喆. 基于深度学习的动态背景下航道船舶检测识别与跟踪研究[D]. 宜昌：三峡大学，2019.

[95]周志华. 机器学习[M]. 北京：清华大学出版社，2016.

[96]黄家才，舒奇，朱晓春，等. 基于迁移学习的机器人视觉识别与分拣策略[J]. 计算机工程与应用，2019, 55(08): 232-237.

[97]庄福振，罗平，何清，等. 迁移学习研究进展[J]. 软件学报，2015, 26(01): 26-39.

[98]王雪松，潘杰，程玉虎. 知识迁移学习方法及应用[M]. 北京：科学出版社，2016.

[99]高随祥，文新，马艳军，等. 深度学习导论与应用实践[M]. 北京：清华大学出版社，2019.

[100]潘格格. 基于深度迁移学习的采煤机摇臂传动系统故障诊断研究[D]. 西安：西安科技大学，2020.

[101]代业滨. 综采工作面机械磨损与润滑失效对策[J]. 设备管理与维修，2021(Z1): 171-174.

[102]杨其明. 磨粒分析：磨粒图谱与铁谱技术[M]. 北京：中国铁道出版社，2002.

[103]LeCun Y, Bottou L, Bengio Y, et al. Gradient-based learning Applied to document recognition[J]. Computer Science, 1998, 86(11): 2278-2324.

[104]Alex K, Ilya S, Geoffrey E H. ImageNet classification with deep convolutional neural networks[J]. 2017, 60(6): 84-90.

[105]Simonyan K, Zisserman A. Very deep convolutional networks for large-scale image recognition[C]. International Conference on Learning Representations, 2015.

[106]Christian S, Liu W, Jia Y Q, et al. Going deeper with convolutions[C]. 2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2015, 1: 1-9, DOI: 10.1109/CVPR.2015.7298594.

[107]He K, Zhang X, Ren S, et al. Deep Residual Learning for Image Recognition[C]. 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2016, 1: 770-778.

Jin Y F, Yin Z Y, Zhou W H, et al. A single objective EPR based model for creep index of soft clays considering L2 regularization[J]. Engineering Geology, 2019, 248: 242-255.