Russian Federation
Russian Federation
TI “VTB”
Russian Federation
Russian Federation
UDC 656.25
With the growing integration of AI systems in the railway sector, it is becoming essential to ensure their robustness against external interference, particularly subtle alterations in input data. This paper analyses the resilience of various neural network architectures, including ResNet18, ResNet50,VisionTransformer, a convolutional neural network, and the GPT-4o multimodal model, which are used for automated fault detection in the analysis of railway infrastructure images. Experiments have been conducted using adversarial disturbances generated via universal noise derived from a set of transformers. Two attack modifications were employed to simulate real-world interference scenarios with limited data availability. The performance of the models has been evaluated on both untainted images and those compromised by overlaid noise. The outcomes indicate that while ResNet50attains maximum accuracy on unmodified data, ViT and GPT-4o demonstrate greater resilience to adversarial disturbances. The research emphasizes the importance of selecting model architectures based on both their accuracy and their robustness against distortions. This paper introduces a method for assessing robustness and offers practical recommendations for developing AI systems designed for application in railway environments where safety is critical.
AI systems, railway transport, adversarial attacks, robustness, neural networks, computer image, safety, infrastructure monitoring
1. Kulagin M. A. Obobschenie opyta resheniya zadach prediktivnoy analitiki na zheleznodorozhnom transporte / M. A. Kulagin, V. G. Sidorenko // Nauka i tehnika transporta. — 2024. — № 4. — S. 55–62.
2. Sidorenko V. G. Intellektual'naya sistema obnaruzheniya narusheniy v soblyudenii trebovaniy bezopasnosti pri rabotah na ob'ektah zheleznodorozhnoy infrastruktury / V. G. Sidorenko, M. A. Kulagin, D. M. Rodina // Avtomatika na transporte. — 2025. — T. 11. — № 1. — S. 55–65. — DOI:https://doi.org/10.20295/2412-9186-2025-11-01-55-65.
3. Malinskiy S. V. Avtomaticheskoe opredelenie granic opasnyh uchastkov zheleznodorozhnogo puti / S. V. Malinskiy, A. V. Abramov, V. O. Sharova // Intellektual'nye transportnye sistemy: materialy IV Mezhdunarodnoy nauchno-prakticheskoy konferencii, Moskva, 22 maya 2025 goda. — Moskva: Rossiyskiy universitet transporta (MIIT), 2025. — S. 624–630. — DOI: 10.30932/ 9785002587582-2025-624-630.
4. Ashrafzyanov A. M. Obnaruzhenie i raspoznavanie pre- pyatstviy pered avtomobilem na osnove obrabotki videoizobrazheniy / A. M. Ashrafzyanov, M.Shleymovich // Vestnik Kazanskogo gosudarstvennogo tehnicheskogo universiteta im. A.N. Tupoleva. — 2014. — № 2. — S. 197–202.
5. Baranov L. A. Metodologiya obosnovaniya trebovaniy bezopasnosti pri ispol'zovanii sistem tehnicheskogo zreniya v intellektual'nyh sistemah upravleniya dvizheniem poezdov / L. A. Baranov, P. F. Bestem'yanov, E. P. Balakina, A. L. Ohotnikov // Intellektual'nye transportnye sistemy: materialy Mezhdunarodnoy nauchno-prakticheskoy konferencii, Moskva, 26 maya 2022 goda. — M.: Rossiyskiy universitet transporta, 2022. — S. 54–58.
6. Ohotnikov A. L. Proekty sistem tehnicheskogo zreniya dlya avtomaticheskogo upravleniya dvizheniem / A. L. Ohotnikov // Avtomatika, svyaz', informatika. — 2023. — № 3. — S. 21–24. — DOI:https://doi.org/10.34649/AT.2023.3.3.003.
7. Ozerov A. V. Tehnicheskoe zrenie v sostave sistem avtomaticheskogo upravleniya dvizheniem poezdov / A. V. Ozerov, A. S. Marshova // Perspektivnye informacionnye tehnologii (PIT 2022): trudy Mezhdunarodnoy nauchno-tehnicheskoy konferencii, Samara, 18–21 aprelya 2022 goda / pod red. S. A. Prohorova. — Samara: Izdatel'stvo Samarskogo nauchnogo centra RAN, 2022. — S. 201–205.
8. Lisanti G. A Multi-Camera Image Processing and Visualization System for Train Safety Assessment / G. Lisanti, S. Karaman, D. Pezzatini // arXiv preprint. — 2015. — arXiv:1507.07815. — DOI:https://doi.org/10.48550/arXiv.1507.07815.
9. Saritas M. M. et al. Railway Track Fault Detection with ResNet Deep Learning Models // 2023 International Conference on Intelligent Systems and New Applications (ICISNA’23). — 2023.
10. Sun' H. Obzor sovremennyh sistem tehnicheskogo zreniya, primenyaemyh v transportnoy otrasli / H. Sun', S. chzhuan, A. A. Kostrov // Sovremennye naukoemkie tehnologii. — 2024. — № 9. — S. 69–73. — DOI: 10.17513/ snt.40150.
11. Goodfellow I. Explaining and Harnessing Adversarial Examples / I. Goodfellow, J. Shlens, C. Szegedy // arXiv preprint. — 2015 (submitted 20 Dec 2014, revised 20 Mar 2015). — arXiv:1412.6572. — DOI:https://doi.org/10.48550/arXiv.14.6572.
12. Grigorenko A. G. Obzor metodov zaschity ot adversial'noy ataki One Pixel v sistemah mashinnogo obucheniya / A. G. Grigorenko, N. A. Vasil'ev, D. S. Sitdikov // Sistemy intellektual'nogo upravleniya i iskusstvennyy intelekt: teoriya i praktika: sbornik trudov II nacional'-noy nauchno-prakticheskoy konferencii, Sankt- Peterburg, 27 iyunya 2024 goda. — SPb.: Federal'noe gosudarstvennoe byudzhetnoe obrazovatel'noe uchrezhdenie vysshego obrazovaniya Gosudarstvennyy universitet morskogo i rechnogo flota im. admirala S. O. Makarova, 2024. — S. 30–36.
13. H'yuber P. Robastnaya statistika / P. H'yuber. M.: Mir,1984.
14. Goodfellow I. J. Making machine learning robust against adversarial inputs / I. J. Goodfellow, J. Shlens, C. Szegedy // Communications of the ACM. — 2018. — Vol. 61. — Iss. 7. — Pp. 56–66. — DOI:https://doi.org/10.1145/3134599.
15. Vasilenko, M. S. Algoritm mashinnogo obucheniya dlya detektirovaniya vybrosov i anomaliy / M. S. Vasilenko, A. S. Kopyrin // Modeling of Artificial Intelligence. — 2019. — № 6-1. — S. 13–18. — DOI:https://doi.org/10.13187/mai.2019.1.13.
16. Legashev L. V. Metodika postroeniya ustoychivoy sistemy zaschity na osnove sostyazatel'nogo mashinnogo obucheniya v besprovodnyh setyah 6G / L. V. Legashev, L. S. Grishina // Voprosy kiberbezopasnosti. — 2023. — № 2(54). — S. 99–108. — DOI:https://doi.org/10.21681/2311-3456-2023-2-99-108.
17. Hou X. High-Speed Rail Operating Environment Recognition Based on Neural Network and Adversarial Training / X. Hou et al. //2019 IEEE 31st International Conference on Tools with Artificial Intelligence (ICTAI). — IEEE, 2019. — Pp. 840– 847. — DOI:https://doi.org/10.1109/ICTAI.2019.00120.
18. Goldobin I. A. Vliyanie shumov na algoritmy cifrovoy obrabotki izobrazheniy / I. A. Goldobin, E. I. Klimova // Aktual'nye voprosy razvitiya sovremennoy cifrovoy sredy: sbornik statey po materialam nauchno-tehnicheskoy konferencii molodyh uchenyh, Moskva, 14–16 aprelya 2021 goda. — Volgograd: Sirius, 2021. — S. 396–402.
19. Kotenko I. V. Ataki i metody zaschity v sistemah mashinnogo obucheniya: analiz sovremennyh issledovaniy / I. V. Kotenko, I. B. Saenko, O. S. Lauta i dr. // Voprosy kiberbezopasnosti. — 2024. — № 1(59). — S. 24–37. — DOI:https://doi.org/10.21681/2311-2024-1-24-37.
20. Kostyumov V. V. Obzor i sistematizaciya atak ukloneniem na modeli komp'yuternogo zreniya / V. V. Kostyumov // International Journal of Open Information Technologies. — 2022. — T. 10. — № 10. — S. 11–20.
21. Potapov A. K. Vulnerabilities of Artificial Intelligence Systems / A. K. Potapov, V. G. Sidorenko // 2024 International Conference “Quality Management, Transport and Information Security, Information Technologies” (QM&TIS&IT). — IEEE, 2024. — Pp. 84–87. — DOI: 10.1109/ QMTISIT63393.2024.10762915.
22. Grachev Ya. L. Ispol'zovanie kachestvennyh harakteristik izobrazheniya dlya kompleksnogo stegoanaliza / Ya. L. Grachev, V. G. Sidorenko // Nadezhnost'. — 2025. —T. 25. — № 1. — S. 67–74. — DOI:https://doi.org/10.21683/1729-2646- 2025-25-1-67-74.
23. Orlov S. P. Glubokaya neyronnaya set' dlya diagnostiki elementov zheleznodorozhnogo rel'sovogo puti / S. P. Orlov, N. A. Efimushkin, N. V. Efimushkina // Vestnik Samarskogo gosudarstvennogo tehnicheskogo universiteta. Seriya: Tehnicheskie nauki. — 2022. — T. 30. — № 1(73). — S. 63–74. — DOI:https://doi.org/10.14498/tech.2022.1.4.
24. Fedorov V. A. Obnaruzhenie zheleznodorozhnoy infrastruktury na osnove YOLOv8 s ispol'zovaniem neyronnogo processora / V. A. Fedorov // Doklady Rossiyskoy akademii nauk. Matematika, informatika, processy upravleniya. — 2024. — T. 520. — № S2. — S. 49–56. — DOI:https://doi.org/10.31857/S2686954324700371.
25. Eunus S. I. ECARRNet: An Efficient LSTM-Based Ensembled Deep Neural Network Architecture for Railway Fault Detection / S. I. Eunus, S. Hossain, A. E. M. Ridwan, A. Adnan et al. // AI. — 2024. — Vol. 5. — Iss. 2. — Pp. 482–503. — DOI:https://doi.org/10.3390/ai5020024.
26. Goodfellow I. Deep Learning / I. Goodfellow, Y. Bengio, A. Courville. — Cambridge, MA: MIT Press, 2016. — DOI:https://doi.org/10.1007/s10710-017-9314-z.
27. Han K. A Survey on Visual Transformer / K. Han, Y. Wang,H. Chen // arXiv preprint. — 2020. — DOI: 10.48550/ arXiv.2012.12556.
28. Powers D. M. W. Evaluation: From Precision, Recall and F-Measure to ROC, Informedness, Markedness & Correlation / D. M. W. Powers // Journal of Machine Learning Technologies. — 2011. — Vol. 2. — Iss. 1. — Pp. 37–63. — DOI:https://doi.org/10.48550/arXiv.2010.16061.
