基於YOLOv4的軌道缺失扣件偵測系統

A YOLOV4 BASED CLOUD SERVER FOR RAILWAY FAULT FASTENER DETECTION

10.6652/JoCICHE.202106_33(4).0002

謝禎冏(Chen-Chiung Hsieh)；林雅雯(Ya-Wen Lin)；黃維信(Wei-Hsin Huang)；謝尚琳(Shang-Lin Hsieh)；洪瑋宏(Wei-Hung Hung)

軌道扣件巡檢 ； 自動缺失扣件偵測 ； 深度學習 ； YOLOv4 ； track fastener inspection ； automatic fault fastener detection ； deep learning ； YOLOv4

中國土木水利工程學刊

33卷4期（2021 / 06 / 01）

263 - 272

繁體中文

鐵路的每支枕木上有4個軌道扣件，用以固定鋼軌於枕木上，現行檢查方式是於夜間以人力目視，在光照不足與長時間的單調工作下，巡檢效率不彰，因此需要自動化的軌道扣件缺失辨識系統。本研究採用高速攝影機使巡檢能夠快速進行，再結合深度學習的目標檢測網路YOLOv4，提供近似於人的軌道扣件缺失狀況判斷。在解析度416 × 416下使用7,300張相片進行訓練，於測試集中的824張相片辨識結果，類別平均準確率、檢出率、與準確率分別達到91.52%、86%、和84%。在伺服器搭配GPU 2080Ti下，辨識速度可達31.25 FPS。以其他區段實地驗證，大甲站至臺中港站來回共四趟，本系統檢出率為88%，且偵測出比Ground Truth更多的缺失扣件，證實本系統實用性。

There are 4 track fasteners on each sleeper of the railway to fix the track on the sleeper. The current inspection method is to use human visual inspection at night. Under the lack of light and long-term monotonous work, the inspection efficiency is not good. Therefore, an automatic fault detection system of track fastener is in urgent need. In this study, high-speed cameras are used to make inspections fast, classified by the deep learning neural network model YOLOv4, to provide a judgment of the fault track fasteners similar to human's. After training with 7,300 images of resolution 416 × 416, the recognition results of the test set consisting 824 images, the mAP, recall, and accuracy reached 91.52%, 86%, and 84%, respectively. Under the server with CPU AMD 2700 and GPU 2080Ti, the recognition speed can reach 31.25 FPS. Four round-trips field verification in other regions from Dajia Station to Taichung Port Station were conducted. The system's recall rate was 88%, and more fault fasteners than ground truth were detected, confirmed the practicality of the system.

1. 林宏達(2018).光學非接觸式軌道檢測分析設備於軌道管理之應用.土木水利,45(1),106-118.
   連結：
2. “How Swiss Federal Railways is improving passenger safety with the power of deep learning,” Forbes (2018).
3. Bochkovskiy, A.,Wang, C. Y.,Mark Liao, H.-Y.(2020).,未出版
4. bvSys Bildverarbeitungssysteme, “Visual inspection systems for track maintenance,” https://www.railwaytechnology.com/contractors/track/bvsys/，查詢日期：2020 年 6 月 18 日。
5. Cui, H.,Li, J.,Hu, Q.,Mao, Q.(2019).Real-time inspectionsystem for ballast railway fasteners based on point cloud deep learning.IEEE Access,8,61604-61614.
6. ENSCO, “Track imaging systems,” https://www.ensco.com/rail/track-imaging-systems，查詢日期：2020 年 1 月19 日。
7. Hsieh, C.-C.,Lin, Y.-W.,Tsai, L.-H.,Huang, W.-H.,Hsieh, S.-L.,Hung, W.-H.(2020).Offline deep-learning based defective track fastener detection and inspection system.Sensors and Materials,32(10),3429-3442.
8. Huang, Z.,Wang, J.(2019).,未出版
9. Lin, Y.-W.,Hsieh, C.-C.,Huang, W.-H.,Hsieh, S.-L.,Hung, W.-H.(2019).Railway track fasteners fault detection using deep learning.IEEE Eurasia Conference on IOT, Communication and Engineering
10. Liu, Wei,Anguelov, Dragomir,Erhan, Dumitru,Szegedy, Christian,Reed, Scott,Fu, Cheng Yang,Berg, Alexander C.(2016).SSD: Single Shot multibox Detector.Lecture Notes in Computer Science
11. Maj, Michal, “Object detection and image classification with YOLO,” Sep. 2018. [Online]. Available: https://tinyurl.com/ycq7wkm9
12. Min, Y.,Xiao, B.,Dang, J.,Yue, B.,Cheng, T.(2018).Real time detection system for rail surface defects based on machine vision.EURASIP Journal on Image and Video Processing,3,11.
13. Pavemetrics, “Laser Rail Inspection System (LRAIL),” Canada (2019), http://www.pavemetrics.com/applications/ rail-inspection/laser-rail-inspection-system/,查詢日期：2019 年 1 月 14 日。
14. Redmon, J.,Farhadi, A.(2018).,未出版
15. Ren, Shaoqing,He, Kaiming,Girshick, Ross,Sun, Jian(2015).Faster R-CNN: Towards real-time object detection with region proposal networks.Neural Information Processing Systems
16. Song, Q.,Guo, Y.,Jiang, J.,Liu, C.,Hu, M.(2019).,未出版
17. Synodinos, A. P.(2016).Identification of railway track components and defects by analysis of wheel-rail interaction noise.23rd International Congress on Sound & Vibration
18. Tan, M.,Pang, R.,Le, Q. V.(2020).EfficientDet: Scalable and efficient object detection.IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR),Seattle, WA, USA:
19. Wang, T.,Yang, F.,Tsui, K-L.(2020).Real-time detection of railway track component via one-stage deep learning networks.Sensors,20(15),4325.
20. Wei, X.,Yang, Z.,Liu, Y.,Wei, D.,Jia, L.,Li, Y.(2019).Railway track fastener defect detection based on image processing and deep learning techniques: A comparative study.Engineering Applications of Artificial Intelligence,80,66-81.
21. 台北捷運公司，「TRTC 臺北捷運北投機廠軌道檢查車 EM-80E 」 ， https://www.youtube.com/watch?v= JqEQARYCkag，查詢日期：2020 年 6 月 18 日。
22. 朱旭，「軌道工程」，取自 http://www.ct.ntust.edu.tw/ ct/files/files/2008-04-14-03.pdf，查詢日期：2021 年 2 月25 日。
23. 陳嘉昌,楊子賸,王怡華,黃彥斌,林安祥(2014)。軌道扣夾之功能要求與設計測試。臺鐵資料季刊
24. 謝禎冏,林雅雯,黃維信,謝尚琳,洪瑋宏,李明德(2020)。扣件缺失辨識系統建置研究。交通部運輸研究所港灣季刊，第一季，20-32。