Intelligent video analysis model for large-scale construction machinery supervision system
-
摘要:
为在铁路工程中加强对铁路大型施工机械的安全管理，帮助建设管理单位实现对施工现场大型施工机械的整体掌控，设计了大型施工机械监管系统，介绍了其总体架构，并重点阐述了其中智能视频分析模型的设计。该模型基于YOLOv6模型，结合迁移学习、不平衡学习、数据增强等多种深度学习技术，实现铁路大型施工机械的快速定位与分类。模型的宏平均准确率可达94.0%、mAP可达0.956、每秒检测帧数可达84，准确性和实时性均满足实际应用需求。
Abstract:In order to strengthen the safety management of large-scale construction machinery in railway engineering and help construction management units achieve overall control of large-scale construction machinery on construction sites, this paper designed a large-scale construction machinery supervision system, introduced its overall architecture, and focused on the design of an intelligent video analysis model. This model was based on the YOLOv6 model, combined with various deep learning techniques such as transfer learning, imbalanced learning, and data augmentation, to implement rapid positioning and classification of large-scale railway construction machinery. The average macro accuracy of the model can reach 94.0%, mAP can reach 0.956, and the detection frame rate per second can reach 84. Its accuracy and real-time performance meet practical application requirements.
-
-
表 1 训练环境
环境配置 名称 配置 硬件配置 GPU NVIDIA GeForce GTX 1080 Ti CPU Intel（R） Xeon（R） CPU E5-2650 v4 内存 16 G 显存 12 G 软件配置 操作系统 Linux Python 3.8.0 Pytorch 1.8.0 CUDA 11.1 cuDNN 8.1.0 
下载: 导出CSV
表 3 不同模型训练方法的效果对比
处理方法 macro-ACC macro-R macro-P mAP 无 31.3% 37.4% 55.2% 0.129 +迁移学习 70.5% 85.7% 73.1% 0.537 +迁移学习+数据增强 81.5% 97.9% 83.1% 0.769 +迁移学习+数据增强+不平衡学习 94.0% 98.3% 95.7% 0.956
下载: 导出CSV
-
[1] 刘祥敏. 临近铁路营业线大型机械设备的施工安全监管[J]. 设备管理与维修,2017(9):23-25. DOI: 10.16621/j.cnki.issn1001-0599.2017.07D.11. [2] 朱涨鑫,谢以顺,铁 栋,等. 基于UWB的铁路营业线施工要素定位与风险防控研究[J/OL]. 铁道标准设计:1-8[2023-11-14]. https://doi.org/10.13238/j.issn.1004-2954.202305020002. [3] 徐 鑫,潘 杰,曹利安,等. 基于深度学习的铁路异物侵限检测模型[J]. 铁路计算机应用,2023,32(10):7-12. DOI: 10.3969/j.issn.1005-8451.2023.10.02. [4] Tan C Q, Sun F C, Kong T, et al. A survey on deep transfer learning[C]//27th International Conference on Artificial Neural Networks and Machine Learning, 4-7 October, 2018, Rhodes, Greece. Cham, Switzerland: Springer, 2018: 270-279.
[5] Ribani R, Marengoni M. A survey of transfer learning for convolutional neural networks[C]//2019 32nd SIBGRAPI Conference on Graphics, Patterns and Images Tutorials (SIBGRAPI-T), 28-31 October, 2019, Rio de Janeiro, Brazil. New York, USA: IEEE, 2019: 47-57.
[6] 周 玉,孙红玉,房 倩,等. 不平衡数据集分类方法研究综述[J]. 计算机应用研究,2022,39(6):1615-1621. DOI: 10.19734/j.issn.1001-3695.2021.10.0590. [7] Singh J, Beeche C, Shi Z Y, et al. Batch-balanced focal loss: a hybrid solution to class imbalance in deep learning[J]. Journal of Medical Imaging, 2023, 10(5): 051809.
-
期刊类型引用(2)
1. 李丽辉，王炜炜，张涛，景辉. 面向冬奥的京张高速铁路智能票务技术体系研究. 铁道运输与经济. 2022(09): 19-25 . 
百度学术
2. 马建军. 京张高速铁路智能化服务关键技术研究与冬奥科技保障应用示范. 铁道运输与经济. 2022(09): 1-10 . 百度学术
其他类型引用(0)
计量
- 文章访问数: 105
- HTML全文浏览量: 29
- PDF下载量: 54
- 被引次数: 2
