Determination of the optimal neural network architecture for the search for the railway fishplates

Authors: Marenov N.E.
Published in issue: #6(47)/2020
DOI: 10.18698/2541-8009-2020-6-618
Category: Instrument Engineering, Metrology, Information-Measuring Instruments and Systems | Chapter: Laser and opto-electronic systems
Keywords: convolutional neural network, deep learning, machine learning, image classification, pattern recognition, rail, fishplates, defect detection
Published: 21.06.2020

The use of deep neural networks is currently one of the most popular approaches to creating automated systems. Convolutional neural networks are used in image classification problems. Automation of monitoring the condition of the railway track is necessary for the smooth movement of transport, therefore, the solution of the problem of detecting defects using neural networks is relevant. The article defines the optimal architecture of the neural network for the search for fishplates, consisting of seven layers. The influence of the size of the submitted images and the number of Convolution – ReLU – Max Pooling layers on the image processing speed and classification errors is considered.

References

[1] Yakovleva T.G., ed. Zheleznodorozhnyy put’ [Railway]. Moscow, Transport Publ., 1999. (in Russ.).

[2] Shakhunyants G.M. Zheleznodorozhnyy put’ [Railway]. Moscow, Transport Publ., 1987. (in Russ.).

[3] Sadeghi J., Askarinejad H. Application of neural networks in evaluation of railway track quality condition. J. Mech. Sci. Technol., 2012, vol. 26, no. 1, pp. 113–122. DOI: https://doi.org/10.1007/s12206-011-1016-5

[4] Nakhaee M.Ch., Hiemstra D., Stoelinga M., et al. The recent applications of machine learning in rail track maintenance: a survey. Proc. RSSRail, 2019, pp. 91–105. DOI: https://doi.org/10.1007/978-3-030-18744-6_6

[5] Gibert X., Patel V., Chellappa R. Deep multitask learning for railway track inspection. IEEE Trans. Intell. Transp. Syst., 2017, vol. 18, no. 1, pp. 153–164. DOI: https://doi.org/10.1109/TITS.2016.2568758

[6] Mittal Sh., Rao D. Vision based railway track monitoring using deep learning. arxiv.org: website. URL: https://arxiv.org/abs/1711.06423 (accessed: 15.02.2020).

[7] Nikolenko S., Kadurin A., Arkhangel’skaya E. Glubokoe obuchenie [Deep learning]. Sankt-Petersburg, Piter Publ., 2019 (in Russ.).

[8] Svertochnye neyronnye seti: vzglyad iznutri [Convolutional neural networks: an outward glance]. ru.datasides.com: website (in Russ.). URL: http://ru.datasides.com/code/cnn-convolutional-neural-networks/ (accessed: 16.12.2019).

[9] Cawley G., Talbot N. On over-fitting in model selection and subsequent selection bias in performance evaluation. J. Mach. Learn. Res., 2010, no. 11, pp. 2079–2107.

[10] Hastie T., Tibshirani R., Friedman J. The elements of statistical learning. Springer, 2009.