Defect detection method for track fastener based on Transformer and local feature fusion
-
摘要:
为解决传统人工巡检轨道交通线路存在的效率低和有安全隐患等问题，提出一种基于Transformer与局部特征融合的轨道紧固件缺陷检测方法。构建轨道紧固件缺陷检测模型，将Transformer与局部特征模块融合，整合局部信息，进而提取轨道紧固件缺陷特征；同时，采用数据增强的方法对轨道紧固件缺陷样本进行数据扩增，扩充数据集，验证所建模型的检测效果。实验结果表明，相较于传统方法，文章提出的方法在识别轨道紧固件缺失和损坏两类缺陷方面的精度和平均准确率均有所提升，在不同的轨道线路实验环境下也表现出良好的检测效果。
-
关键词:
- 轨道线路 /
- 紧固件缺陷检测 /
- Transformer /
- 局部特征 /
- 数据增强
Abstract:To solve the problems of low efficiency and safety hazards in traditional manual inspection of rail transit lines, this paper proposed a rail fastener defect detection method based on Transformer and local feature fusion. The paper constructed a defect detection model for rail fasteners, integrated Transformer with local feature modules, integrated local information, and extracted defect features of rail fasteners, at the same time, used data augmentation methods to expand the dataset of rail fastener defect samples and verify the detection effect of the constructed model. The experimental results show that compared to traditional methods, the proposed method has improved accuracy and average accuracy in identifying two types of defects, namely missed and damaged track fasteners. It also shows good detection performance in different track experimental environments.
-
Keywords:
- track line /
- fastener defect detection /
- Transformer /
- local features /
- data enhancement
-
-
表 1 实验环境
配置 型号 实验平台 Ubuntu18.04.6LTS 深度学习框架 PyTorch CPU Intel（R） Core（TM） i9-9900KF CPU @ 3.60GHz GPU NVIDIA TITAN RTX 编程语言 Python 3.8.17 
下载: 导出CSV
表 2 模型参数设置
参数 设置 图像大小 384 × 384 优化算法 Adam 动量系数 0.9 学习率 0.0001 权重衰减系数 0.01 batch size 32 注意力头数量 6
下载: 导出CSV
表 3 4种方法的检测结果对比
方法 P R mAP ResNet-50 84.8% 90.3% 80.2% YOLOv3 87.8% 90.6% 81.9% ViT 89.5% 92.3% 84.7% 本文方法 91.4% 96.9% 86.1%
下载: 导出CSV
-
[1] Liu S, Wang Q D, Luo Y P. A review of applications of visual inspection technology based on image processing in the railway industry[J]. Transportation Safety and Environment, 2019, 1(3): 185-204. DOI: 10.1093/tse/tdz007
[2] 卢艳东. 基于深度学习的轨道紧固件检测算法研究[D]. 兰州:兰州交通大学,2022. [3] 周 颖. 基于深度学习的轨道扣件缺陷检测方法研究[D]. 成都:四川大学,2021. [4] Liu J B, Huang Y P, Zou Q, et al. Learning visual similarity for inspecting defective railway fasteners[J]. IEEE Sensors Journal, 2019, 19(16): 6844-6857. DOI: 10.1109/JSEN.2019.2911015
[5] 马 茜. 基于图像识别技术的轨道交通缺陷检测研究[J]. 计算技术与自动化,2022,41(1):117-122. [6] Wang T G, Zhang Z J, Yang F F, et al. Automatic rail component detection based on AttnConv-net[J]. IEEE Sensors Journal, 2022, 22(3): 2379-2388. DOI: 10.1109/JSEN.2021.3132460
[7] 王悦林. 基于BERT的对AI理解语言方式的研究[J]. 科技视界,2019(5):88-89. [8] 温忠凯. 铁路钢轨扣件发展综述[J]. 商品与质量,2015(19):224-225. [9] 程智余,张金锋,孙丙宇. 基于Transformer和注意力机制的角钢塔螺栓缺陷检测模型[J]. 计算机系统应用,2023,32(4):248-254. [10] 唐东林,杨 洲,程 衡,等. 浅层卷积神经网络融合Transformer的金属缺陷图像识别方法[J]. 中国机械工程,2022,33(19):2298-2305,2316. [11] Vaswani A, Shazeer N, Parmar N, et al. Attention is all you need[C]//Proceedings of the 31st International Conference on Neural Information Processing Systems, 4 December, 2017, Long Beach, California, USA. Red Hook, USA: Curran Associates Inc. , 2017: 6000-6010.
[12] 刘 畅,莫海芳,马 春. 基于Transformer和CNN的真实场景下植物病害识别方法[J]. 现代计算机,2023,29(11):22-27. [13] Wang F, Jiang M Q, Qian C, et al. Residual attention network for image classification[C]//Proceedings of 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 21-26 July, 2017, Honolulu, HI, USA. New York, USA: IEEE, 2017: 6450-6458.
[14] 安小松,宋竹平,梁千月,等. 基于CNN-Transformer的视觉缺陷柑橘分选方法[J]. 华中农业大学学报,2022,41(4):158-169. -
期刊类型引用(0)
其他类型引用(1)
计量
- 文章访问数: 59
- HTML全文浏览量: 30
- PDF下载量: 39
- 被引次数: 1
