THE ELECTRONIC SCIENTIFIC JOURNAL "YOUNG SCIENCE OF SIBERIA"

SOFTWARE IMPLEMENTATION OF THE UPPER STRUCTURE OF THE RAILWAY TRACK DEFECTS DETECTION AUTOMATED SYSTEM BASED ON THE TECHNOLOGY OF THE CONVOLUTIONAL NEURAL NETWORKS

Receipt date: 
24.10.2018
Bibliographic description of the article: 

Reznitskiy M.A., Arshinsiy L.V. Programmnaya realizatsiya avtomatizirovannoi systemi obnaryjeniya defectov verkhnego cttroeniya puti na osnove tekhnologii svertochnykh neyronnykh setey [Software implementation of an automated system for detecting defects of the upper structure of the railway path based on the technology of convolutional neural networks]. Molodaya nauka Sibiri: ehlektronnyj nauchnyj zhurnal [Young science of Siberia: electronic scientific journal], 2018, no. 1. [Accessed 24/10/18]

Year: 
2018
Journal number: 
УДК: 
004.8
Article File: 
Abstract: 

In the article, based on the technology of convolutional neural networks, approaches to solving the problem of automated processing of images of the upper structure of the railway track in order to identify areas suspicious of defects are considered. Such defects are considered as defects, rail joints and defects, rail fasteners, sleepers defects, surface defects in rails, "other", which include all the defects, not included in the previous classification. At the same time, due to the high cost of the second kind of error, all images with the help of the network are divided into two classes: images without defects and images that are suspicious of defects. The second set of images is proposed to be sent to specialists for manual verification. The source data for the image processing process are files with snapshots of the paths. Each photo represents a shift relative to the previous one by about 30 cm along the paths. The images received by one of trains of the center of diagnostics and monitoring of devices of infra-structure – structural division of the West Siberian management of infrastructure-structural division of the West Siberian railway – branch of JSC RZHD were taken as a basis. The content and features of the method and its software implementation are discussed. The software package includes a program for automated image segmentation, a program for automated Assembly of training and test data sampling into a text file, a program for configuration and training of a convolutional neural network. The principal possibility of using convolutional neural networks for solving the given class of problems is shown.

List of references: 

1. Boravskaya E.N., Shapilov E.D. Predposylki dlya formirovaniya mezhdunarodnoy seti VSM [Prerequisites for the formation of an international network of HSR] // Skorostnoy i vysokoskorostnoy zheleznodorozhnyy transport. 2001. V. 1. pp. 181-183. (in Russian)

2. Borisov E.S. O metodakh obucheniya mnogosloynykh neironnykh setey pryamogo rasprostraneniya [Methods of training multilayered neural networks of direct distribution] – http://mechanoid.kiev.ua/neural-net-backprop2.html (data obrasheniya 25.09.2018) (in Russian)

3. Reznitskiy M.A. Resheniye zadachi analiza verkhnego stroeniya puti po ego izobrazeniyam [Solving the problem of analysis of the upper structure of the railway path by its images]. Informatsionnyye tekhnologii i problemy matematicheskogo modelirovaniya slozhnykh sistem. Vol. 16. Irkutsk: IrGUPS, 2016. pp 15-20. (in Russian)

4. Phisenko V.T., Phisenko T.Yu. Kompyuternaya obrabotka i raspoznavaniye izobrazheniy [Computer processing and image recognition]. Sankt-Petersburg: SPBGU ITMO, 2008. 281 p. (in Russian)

5. Khaikin S. Neyronnye seti: polnyi kurs, 2-e izd: per. s angl. N.N. Kossul, A.Yu. Shelestova; pod red. N.N. Kossul [Neural networks: full course, 2nd edition: Per. with English. N. N. Kussul, A. Shelestova, Y. ; ed.N. Kussul]. M.: Izdatelskiy dom “Vilyams”, 2006. 1104 p. (in Russian)

6. Hall, E. L. Computer Image Processing and Recognition, Academic Press, New York, 1979. – 811p.

7. Jahne B. Digital image processing. 6th ed. Springer, 2005. – 639 p.

8. Kozin, N.E. Gradual learning the radial neural networks / N.E. Kozin, V.A. Fursov // Computer Optics. – 2004. – № 26. – pp. 138-141.

9. LeCun, Y. Convolutional Networks for Images, Speech, and Time-Series, in Arbib / LeCun, Y. Bengio ; MIT Press, 1995. 14 p.

10. LeCun, Y. Gradient Based Learning Applied to Document Recognition / Y. LeCun, L. Bottou, P. Haffner – IEEE Press, 1998. – P. 46