列车闸瓦钎及闸瓦故障图像检测算法研究

doi:10.11996/JG.j.2095-302X.2023010088

摘要/Abstract

摘要：

列车闸瓦钎及闸瓦状态正常与否对于货运列车安全运行极其重要，为此提出了一种改进的SSD目标检测算法，对货运列车闸瓦钎及闸瓦的缺失进行检测。首先将深度可分离卷积模块引入ResNet50网络模型，使其参数数量减少约50%，以提高检测效率。其次，将改进后的ResNet50网络模型替换SSD中的VGG16网络，以改善SSD网络模型的特征提取能力。然后利用高、低层特征融合的方法，将网络中Conv5_3和Conv7_2分别与Conv4_6和Conv6_2进行特征融合，提高检测精度。最后通过自建的货运列车制动部件缺失数据集对网络进行训练获取比较准确的网络权重。实验结果表明，改进后的SSD算法在闸瓦钎丢失检测中准确率达到96.85%，召回率达到89.50%；在闸瓦丢失检测中准确率达到97.01%，召回率达到97.01%，可以满足列车闸瓦钎及闸瓦缺失检测需求。

关键词: 目标检测, SSD, 深度可分离卷积, 残差结构

Abstract:

The state of brake shoe bolt and brake shoe is of great importance to the safe operation of freight trains. Therefore, an improved SSD (single shot multi-box detector) target detection algorithm was proposed to detect the missing brake shoe bolt and brake shoe of freight trains. Firstly, the depthwise separable convolution module was introduced into the ResNet50 network model to reduce the number of parameters by about 50%, thereby improving the detection efficiency. Secondly, the improved ResNet50 network model was employed to replace the VGG16 network in SSD to improve the feature extraction capability of the SSD network model. Then, Conv5_3 and Conv7_2 were combined with Conv4_6 and Conv6_2 respectively by the method of high level feature and low level feature fusion to improve the detection accuracy. Finally, the network was trained to obtain more accurate weights through the self-built dataset of freight train brake component loss. The experimental results show that the improved SSD algorithm could attain an accuracy of 96.85% and a recall of 89.50% in brake shoe brazing loss detection, and an accuracy of 97.01% and a recall of 97.01% in the detection of brake shoe loss, thus meeting the requirement of missing brake shoe bolt and brake shoe detection.

Key words: target detection, single shot multi-box detector, depthwise separable convolution, residual structure

中图分类号:

TP391

谷雨, 赵军. 列车闸瓦钎及闸瓦故障图像检测算法研究[J]. 图学学报, 2023, 44(1): 88-94.

GU Yu, ZHAO Jun. Research on image detection algorithm of freight train brake shoe bolt and brake shoe fault[J]. Journal of Graphics, 2023, 44(1): 88-94.

图/表 12

参考文献 18

[1]	冯映科. 心盘螺栓和闸瓦钎故障图像检测算法研究[D]. 兰州: 兰州交通大学, 2018.
	FENG Y K. Research on fault image detection algorithm of center plate bolts and brake shoe key[D]. Lanzhou: Lanzhou Jiatong University, 2018 (in Chinese).
[2]	姚炜铭, 王晓华, 吴楠. 基于改进SSD模型的缝纫手势图像检测方法[J]. 激光与光电子学进展, 2020, 57(18): 189-196.
	YAO W M, WANG X H, WU N. Sewing gesture image detection method based on improved SSD model[J]. Laser & Optoelectronics Progress, 2020, 57(18): 189-196 (in Chinese).
[3]	袁帅, 王康, 单义, 等. 基于多分支并行空洞卷积的多尺度目标检测算法[J]. 计算机辅助设计与图形学学报, 2021, 33(6): 864-872.
	YUAN S, WANG K, SHAN Y, et al. Multi-scale object detection method based on multi-branch parallel dilated convolution[J]. Journal of Computer-Aided Design ＆ Computer Graphics, 2021, 33(6): 864-872 (in Chinese).
[4]	SHELHAMER E, LONG J, DARRELL T. Fully convolutional networks for semantic segmentation[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017, 39(4): 640-651. DOI PMID
[5]	GIRSHICK R. Fast R-CNN[C]//2015 IEEE International Confer ence on Computer Vision. New York: IEEE Press, 2015: 1440-1448.
[6]	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 and Machine Intelligence, 2017, 39(6): 1137-1149. DOI PMID
[7]	GIRSHICK R, DONAHUE J, DARRELL T, et al. Rich feature hierarchies for accurate object detection and semantic segmentation[C]//2014 IEEE Computer Society Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2014: 580-587.
[8]	REDMON J, DIVVALA S, GIRSHICK R, et al. You Only Look Once: unified, real-time object detection[C]//2016 IEEE Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2016: 779-788.
[9]	LIU W, ANGUELOV D, ERHAN D, et al. SSD: single shot multibox detector[C]// European Conference on Computer Vision. Cham: Springer International Publishin, 2016: 21-37.
[10]	李兆洋. 基于深度学习的高铁扣件检测算法研究[D]. 成都: 西南交通大学, 2020.
	LI Z Y. Research on high speed rail fastener detection algorithm based on deep learning[D]. Chengdu: Southwest Jiaotong University, 2020 (in Chinese).
[11]	罗隆福, 叶威, 王健. 基于深度学习的高铁接触网顶紧螺栓的缺陷检测[J]. 铁道科学与工程学报, 2021, 18(3): 605-614.
	LUO L F, YE W, WANG J. Defect detection of the puller bolt in high-speed railway catenary based on deep learning[J]. Journal of Railway Science and Engineering, 2021, 18(3): 605-614 (in Chinese).
[12]	赵冰, 代明睿, 李平, 等. 基于深度学习的铁路关键部件缺陷检测研究[J]. 铁道学报, 2019, 41(8): 67-73.
	ZHAO B, DAI M R, LI P, et al. Research on defect detection of railway key components based on deep learning[J]. Journal of the China Railway Society, 2019, 41(8): 67-73 (in Chinese).
[13]	郭忠峰, 张渊博, 王赫莹, 等. 深度学习目标检测算法在货运列车车钩识别中的应用[J]. 铁道科学与工程学报, 2020, 17(10): 2479-2484.
	GUO Z F, ZHANG Y B, WANG H Y, et al. Application of deep learning target detection algorithm in freight train coupler recognition[J]. Journal of Railway Science and Engineering, 2020, 17(10): 2479-2484 (in Chinese).
[14]	汪洋, 王俊刚. 基于深度学习算法的铁路列车运行安全检测[J]. 中国安全科学学报, 2018, 28(S2): 41-45. DOI
	WANG Y, WANG J G. Study on safety inspection of railway train operation based on deep learning algorithm[J]. China Safety Science Journal, 2018, 28(S2): 41-45 (in Chinese). DOI
[15]	刘鑫, 陈思溢, 陈小龙, 等. 基于深度学习的深层次多尺度特征融合目标检测算法[J]. 激光与光电子学进展, 2021, 58(12): 320-328.
	LIU X, CHEN S Y, CHEN X L, et al. Deep multi-scale feature fusion target detection algorithm based on deep learning[J]. Laser & Optoelectronics Progress, 2021, 58(12): 320-328 (in Chinese).
[16]	张云佐, 杨攀亮, 李汶轩. 基于改进SSD算法的铁路隧道漏缆卡具检测[J]. 激光与光电子学进展, 2021, 58(22): 391-398.
	ZHANG Y Z, YANG P L, LI W X. Detection of cable leakage fixture in railway tunnel based on improved SSD algorithm[J]. Laser ＆ Optoelectronics Progress, 2021, 58(22): 391-398 (in Chinese).
[17]	HOWARD A G, ZHU M L, CHEN B, et al. MobileNets: efficient convolutional neural networks for mobile vision applications[EB/OL]. [2021-12-17]. https://arxiv.org/abs/1704.04861.
[18]	汪亚妮, 汪西莉. 基于注意力和特征融合的遥感图像目标检测模型[J]. 激光与光电子学进展, 2021, 58(2): 363-371.
	WANG Y N, WANG X L. Remote sensing image target detection model based on attention and feature fusion[J]. Laser & Optoelectronics Progress, 2021, 58(2): 363-371 (in Chinese).

层数	输出尺寸	结构、类型
Conv 1	150×150	7×7, 64, s₂
Pool 1	75×75	3×3, max, s₂
Conv 2	75×75	$\left( \begin{matrix} 1\times 1, \\ 3\times 3, \\ 1\times 1, \\ \end{matrix} \right.\left. \begin{matrix} 64 \\ 64 \\ \ 512 \\ \end{matrix} \right)\times 3,\ {{\text{s}}_{2}}$
Conv 3	38×38	$\left( \begin{matrix} 1\times 1, \\ 3\times 3, \\ 1\times 1, \\ \end{matrix} \right.\left. \begin{matrix} \ 128 \\ \ 128 \\ \ 512 \\ \end{matrix} \right)\times 4,\ {{\text{s}}_{2}}$
Conv 4	38×38	$\left( \begin{matrix} 1\times 1, \\ 3\times 3, \\ 1\times 1, \\ \end{matrix} \right.\left. \begin{matrix} 256 \\ 256 \\ \ 1024 \\ \end{matrix} \right)\times 6,\ {{\text{s}}_{1}}$
Conv 5	19×19	$\left( \begin{matrix} 1\times 1, \\ 3\times 3, \\ 1\times 1, \\ \end{matrix} \right.\left. \begin{matrix} \begin{matrix} 256 \\ \ 256, \\ 512 \\ \end{matrix} & ds \\ \end{matrix} \right)\times 3,\ {{\text{s}}_{\text{2}}}$

层数	输出尺寸	结构、类型
Conv 1	150×150	7×7, 64, s₂
Pool 1	75×75	3×3, max, s₂
Conv 2	75×75	$\left( \begin{matrix} 1\times 1, \\ 3\times 3, \\ 1\times 1, \\ \end{matrix} \right.\left. \begin{matrix} 64 \\ 64 \\ \ 512 \\ \end{matrix} \right)\times 3,\ {{\text{s}}_{2}}$
Conv 3	38×38	$\left( \begin{matrix} 1\times 1, \\ 3\times 3, \\ 1\times 1, \\ \end{matrix} \right.\left. \begin{matrix} \ 128 \\ \ 128 \\ \ 512 \\ \end{matrix} \right)\times 4,\ {{\text{s}}_{2}}$
Conv 4	38×38	$\left( \begin{matrix} 1\times 1, \\ 3\times 3, \\ 1\times 1, \\ \end{matrix} \right.\left. \begin{matrix} 256 \\ 256 \\ \ 1024 \\ \end{matrix} \right)\times 6,\ {{\text{s}}_{1}}$
Conv 5	19×19	$\left( \begin{matrix} 1\times 1, \\ 3\times 3, \\ 1\times 1, \\ \end{matrix} \right.\left. \begin{matrix} \begin{matrix} 256 \\ \ 256, \\ 512 \\ \end{matrix} & ds \\ \end{matrix} \right)\times 3,\ {{\text{s}}_{\text{2}}}$

数据集	闸瓦钎丢失	闸瓦钎正常	闸瓦丢失	闸瓦钎正常
训练集	981	1359	479	1917
测试集	109	155	51	213
总计	1090	1514	530	2130

数据集	闸瓦钎丢失	闸瓦钎正常	闸瓦丢失	闸瓦钎正常
训练集	981	1359	479	1917
测试集	109	155	51	213
总计	1090	1514	530	2130

特征提取网络	分类准确度(%)
特征提取网络	闸瓦钎丢失	闸瓦丢失
MobileNet	88.00	94.25
ResNet50	88.17	96.18
ResNet50dw	88.33	96.91