Please wait a minute...
Journal of ZheJiang University (Engineering Science)  2021, Vol. 55 Issue (3): 586-593    DOI: 10.3785/j.issn.1008-973X.2021.03.020
    
Metro location point matching and false alarm elimination based on FCM algorithm of target image
Ying-jie ZHENG1,2(),Song-rong WU1,2,Ruo-yu WEI1,2,Zhen-wei TU1,2,Jin LIAO3,Dong LIU1,*()
1. School of Electrical Engineering, Southwest Jiaotong University, Chengdu 611756, China
2. Key Laboratory of Magnetic Suspension Technology and Maglev Vehicle, Ministry of Education, Southwest Jiaotong University, Chengdu 611756, China
3. Sichuan Juzhi Jingchuang Rail Transit Technology Co. Ltd, Chengdu 610000, China
Download: HTML     PDF(899KB) HTML
Export: BibTeX | EndNote (RIS)      

Abstract  

A fuzzy C-means (FCM) algorithm was used to match the location point model in view of the false alarm problem of the track location point detection model, in the process of accurate and fast positioning of metro track based on image recognition. On the basis of the detection model of track location points based on the deep convolution neural network, six kinds of location point images and two kinds of false alarm point images was selected, and then the features (six dimensional feature quantities of each image) such as the center relative position, length width ratio and area of each target detection frame in different types of image samples was extracted. The ReliefF algorithm was adopted to measure the weight value of each dimension feature of all image samples, which was introduced into the Euclidean distance formula of FCM algorithm, so as to uniquely match the location points. Results indicate that the improved FCM algorithm has an obvious improvement in the correctness and effectiveness of clustering, which is of great significance to enhance the accuracy of metro track positioning.



Key wordsimage recognition      deep convolutional neural network      track locating points      fuzzy C-means algorithm      feature weighting     
Received: 18 February 2020      Published: 25 April 2021
CLC:  TP 391.44  
  U 231.94  
Fund:  国家自然科学基金资助项目(61531016);四川省重大科技专项资助项目(20QYCX0095)
Corresponding Authors: Dong LIU     E-mail: 2280511301@qq.com;liudong@swjtu.edu.cn
Cite this article:

Ying-jie ZHENG,Song-rong WU,Ruo-yu WEI,Zhen-wei TU,Jin LIAO,Dong LIU. Metro location point matching and false alarm elimination based on FCM algorithm of target image. Journal of ZheJiang University (Engineering Science), 2021, 55(3): 586-593.

URL:

http://www.zjujournals.com/eng/10.3785/j.issn.1008-973X.2021.03.020     OR     http://www.zjujournals.com/eng/Y2021/V55/I3/586


基于目标图像FCM算法的地铁定位点匹配及误报排除方法

在基于图像识别的地铁轨道精确快速定位过程中,针对轨道定位点检测模型存在的误报问题,将模糊C均值(FCM)算法用于定位点模型匹配. 在基于深度卷积神经网络的轨道定位点检测模型基础上,选用6类定位点图像和2类误报点图像,提取不同类别图像样本中各目标检测框的中心相对位置、长宽比、面积等特征数据(每张图像各6维特征量),采用ReliefF算法度量所有图像样本各维特征量的权重,将所得权重引入FCM算法的欧几里德距离公式,匹配唯一定位点. 实验结果表明,改进后的FCM算法在聚类的正确性和有效性方面有明显改善,对提高地铁轨道定位精度具有重要的意义.


关键词: 图像识别,  深度卷积神经网络,  轨道定位点,  模糊C均值算法,  特征加权 
Fig.1 Target detection process of Darknet-53 model
Fig.2 Flow chart of track location point detection
Fig.3 Picture of location points and false alarm points
样本类别 x y L1 L2 S1 S2
1 1 322 59 0.888 3 0.892 2 150 792 37 128
2 654 ?108 2.848 1 2.233 6 71 100 102 292
3 252 20 0.518 3 0.582 1 98 536 94 068
4 14 ?305 2.837 8 2.824 1 139 860 131 760
5 71 396 0.970 9 0.410 1 41 200 77 252
6 249 379 1.985 5 0.590 2 37 812 140 544
7 72 ?432 0.597 2 3.571 4 49 536 70 000
8 365 90 0.973 3 0.291 1 21 900 52 824
Tab.1 8 sets of standard feature sample data
x y L1 L2 S1 S2
0.241 1 0.291 0 0.286 5 0.254 5 0.260 4 0.225 1
Tab.2 Average weight value of characteristic sample data
Fig.4 Results of thirty runs of ReliefF algorithm
Fig.5 Clustering results of two FCM algorithms
样本类别 u31 u32 u33 u34 u35 u36 u37 u38 u39 u40 u61 u62 u63 u64 u65 u66
1 0.000 8 0.001 2 0.000 3 0.001 0 0.002 4 0.001 7 0.001 5 0.001 9 0.000 2 0.000 6 0.155 2 0.148 3 0.155 0 0.148 0 0.149 5 0.151 1
2 0.001 4 0.001 9 0.000 5 0.001 7 0.004 1 0.003 0 0.002 4 0.003 1 0.000 3 0.001 0 0.148 7 0.152 7 0.152 2 0.149 7 0.149 7 0.148 7
3 0.000 8 0.001 2 0.000 3 0.001 0 0.002 4 0.001 7 0.001 4 0.001 9 0.000 2 0.000 6 0.224 3 0.227 0 0.235 9 0.214 4 0.217 9 0.215 5
4 0.001 1 0.001 6 0.000 4 0.001 3 0.003 3 0.002 2 0.001 8 0.002 5 0.000 2 0.000 7 0.127 4 0.130 2 0.130 3 0.128 0 0.129 3 0.127 3
5 0.096 1 0.094 8 0.058 6 0.152 9 0.193 8 0.716 1 0.255 8 0.258 2 0.030 7 0.920 9 0.083 1 0.083 5 0.078 4 0.088 1 0.086 2 0.086 8
6 0.003 1 0.004 7 0.001 2 0.003 7 0.008 9 0.005 6 0.005 7 0.007 7 0.000 6 0.001 9 0.104 4 0.103 4 0.099 8 0.107 9 0.106 7 0.107 4
7 0.895 8 0.893 6 0.938 3 0.837 4 0.783 0 0.268 1 0.730 0 0.722 8 0.967 8 0.073 7 0.076 8 0.076 9 0.072 5 0.081 1 0.079 5 0.080 1
8 0.000 8 0.001 1 0.000 3 0.001 0 0.002 2 0.001 7 0.001 4 0.001 9 0.000 2 0.000 6 0.080 3 0.078 1 0.075 9 0.082 7 0.081 3 0.083 0
Tab.3 Membership function values of FCM algorithm samples 31~40 and 61~66
算法 CA/% RI
FCM 91.67 0.97
改进的FCM 100.00 1.00
Tab.4 Experimental comparison results of the two FCM algorithms
聚类算法 CA/% RI
K-means算法 77.78 0.93
层次聚类法 91.67 0.89
Tab.5 Experimental comparison results of different clustering algorithms
[1]   翟婉明, 赵春发 现代轨道交通工程科技前沿与挑战[J]. 西南交通大学学报, 2016, 51 (2): 209
ZHAI Wan-ming, ZHAO Chun-fa The frontier and challenge of modern rail transit engineering science and technology[J]. Journal of Southwest Jiaotong University, 2016, 51 (2): 209
[2]   许贵阳, 史天运, 任盛伟, 等 基于计算机视觉的车载轨道巡检系统研制[J]. 中国铁道科学, 2013, 34 (1): 141
XU Gui-yang, SHI Tian-yun, REN Sheng-wei, et al Development of vehicle track inspection system based on computer vision[J]. China Railway Science, 2013, 34 (1): 141
[3]   田贵云, 高斌, 高运来, 等 铁路钢轨缺陷伤损巡检与监测技术综述[J]. 仪器仪表学报, 2016, 37 (8): 1763- 1780
TIAN Gui-yun, GAO Bin, GAO Yun-lai, et al Review of inspection and monitoring technology for rail defects[J]. Chinese Journal of Scientific Instrument, 2016, 37 (8): 1763- 1780
[4]   刘小磊, 黄璞 RFID技术在列车高精度定位中的应用[J]. 都市快轨交通, 2017, 30 (3): 107- 109
LIU Xiao-lei, HUANG Pu Application of RFID technology in high-precision positioning of trains[J]. Urban Rapid Rail Transit, 2017, 30 (3): 107- 109
[5]   赵晓峰 城市轨道交通列车绝对定位系统比较[J]. 城市轨道交通研究, 2015, 18 (10): 57- 58
ZHAO Xiao-feng Comparison of absolute positioning system of urban rail transit trains[J]. Urban Mass Transit, 2015, 18 (10): 57- 58
[6]   路通. 地铁智能巡检车高精度自主定位技术研究[D]. 北京: 北京交通大学, 2019: 7.
LU Tong. Research on high-precision autonomous positioning technology of intelligent inspection vehicle for Metro [D]. Beijing: Beijing Jiaotong University, 2019: 7.
[7]   刘甲甲, 熊鹰, 李柏林, 等 基于计算机视觉的轨道扣件缺陷自动检测算法研究[J]. 铁道学报, 2016, 38 (8): 73- 80
LIU Jia-jia, XIONG Ying, LI Bo-lin, et al Research on automatic detection algorithm of rail fastener defects based on computer vision[J]. Journal of the China Railway Society, 2016, 38 (8): 73- 80
[8]   李志, 陈建政 基于图像处理的铁路轨枕分割方法研究[J]. 科技创新与应用, 2015, (11): 10- 11
LI Zhi, CHEN Jian-zheng Research on railway sleeper segmentation method based on image processing[J]. Technology Innovation and Application, 2015, (11): 10- 11
[9]   张政馗, 庞为光, 谢文静, 等 面向实时应用的深度学习研究综述[J]. 软件学报, 2019, 31 (9): 2655
ZHANG Zheng-kui, PANG Wei-guang, XIE Wen-jing, et al A review of deep learning for real-time applications[J]. Journal of Software, 2019, 31 (9): 2655
[10]   杜馨瑜, 戴鹏, 李颖, 等 基于深度学习的铁道塞钉自动检测算法[J]. 中国铁道科学, 2017, 38 (3): 90
DU Xin-yu, DAI Peng, LI Ying, et al Automatic detection algorithm of railway plug nails based on deep learning[J]. China Railway Science, 2017, 38 (3): 90
[11]   刘军, 阎芳, 杨玺. 物联网与物流管控一体化[M]. 北京: 中国财富出版社, 2017: 431.
[12]   周孟然, 胡锋, 闫鹏程, 等 基于FCM的煤矿突水激光诱导荧光光谱分析[J]. 光谱学与光谱分析, 2018, 38 (5): 1572- 1575
ZHOU Meng-ran, HU Feng, YAN Peng-cheng, et al Analysis of laser-induced fluorescence spectrum of coal mine water inrush based on FCM[J]. Spectroscopy and Spectral Analysis, 2018, 38 (5): 1572- 1575
[13]   杨廷方, 刘沛, 李景禄, 等 FCM结合IEC三比值法诊断变压器故障[J]. 高电压技术, 2007, 33 (8): 66- 69
YANG Ting-fang, LIU Pei, LI Jing-lu, et al FCM combined with IEC three ratio method for transformer fault diagnosis[J]. High Voltage Technology, 2007, 33 (8): 66- 69
[14]   谢福鼎, 李壮 基于改进的半监督FCM算法的高光谱遥感影像分类[J]. 测绘通报, 2016, (9): 60- 63
XIE Fu-ding, LI Zhuang Hyperspectral remote sensing image classification based on improved semi supervised FCM algorithm[J]. Bulletin of Surveying and Mapping, 2016, (9): 60- 63
[15]   肖林云, 陈秀宏, 林喜兰 特征加权和优化划分的模糊C均值聚类算法[J]. 微电子学与计算机, 2016, 33 (10): 143- 145
XIAO Lin-yun, CHEN Xiu-hong, LIN Xi-lan Fuzzy C-Means clustering algorithm for feature weighting and optimal partition[J]. Microelectronics and Computer, 2016, 33 (10): 143- 145
[16]   HE K M, GKIOXARI G, DOLLAR P, et al. Mask R-CNN [C]// IEEE International Conference on Computer Vision. Venice: IEEE, 2017: 2980-2988.
[17]   GIRSHICK R. Fast R-CNN [C]// IEEE International Conference on Computer Vision. Santiago: IEEE, 2015: 1440-1448.
[18]   REN S Q, HE K M, GKIOXARI G, et al Faster R-CNN: towards real-time object detection with region proposal networks[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017, 6 (36): 1137- 1149
[19]   LIU W, ANGUELOV D, ERHAN D, et al. SSD: single shot multibox detector [C]// European Conference on Computer Vision. Amsterdam: Springer, 2016: 21–37.
[20]   REDMON J, FARHADI A. YOLOv3: an incremental improvement [EB/OL]. (2018-04-08) [2020-03-25]. http://arxiv.org/abs/1804.02767.
[21]   何涛, 胡洁, 夏鹏, 等 基于ReliefF算法与遗传算法的肌电信号特征选择[J]. 上海交通大学学报, 2016, 50 (2): 205
HE Tao, HU Jie, XIA Peng, et al Feature selection of EMG signal based on ReliefF algorithm and Genetic algorithm[J]. Journal of Shanghai Jiaotong University, 2016, 50 (2): 205
[22]   刘海洋, 王志海, 张志东 基于ReliefF剪枝的多标记分类算法[J]. 计算机学报, 2019, 42 (3): 489
LIU Hai-yang, WANG Zhi-hai, ZHANG Zhi-dong Multi marker classification algorithm based on ReliefF pruning[J]. Chinese Journal of Computers, 2019, 42 (3): 489
[23]   蒋玉娇, 王晓丹, 王文军 等. 一种基于PCA和ReliefF的特征选择方法[J]. 计算机工程与应用, 2010, 46 (26): 171
JIANG Yu-jiao, WANG Xiao-dan, WANG Wen-jun, et al A feature selection method based on PCA and ReliefF[J]. Computer Engineering and Application, 2010, 46 (26): 171
[24]   高新波, 裴继红, 谢维信 模糊c-均值聚类算法中加权指数m的研究 [J]. 电子学报, 2000, 28 (4): 80
GAO Xin-bo, PEI Ji-hong, XIE Wei-xin Research on weighted index m in fuzzy c-Means clustering algorithm [J]. Journal of Electronics, 2000, 28 (4): 80
[1] Xiao-feng FU,Li NIU,Zhuo-qun HU,Jian-jun LI,Qing WU. Deep micro-expression spotting network training based on concept of transition frame[J]. Journal of ZheJiang University (Engineering Science), 2020, 54(11): 2128-2137.
[2] HUANG Feng, BO Jia-Dun, CHEN Chun, et al. Improving question classification via weighted feature model[J]. Journal of ZheJiang University (Engineering Science), 2009, 43(6): 994-998.