Powered by OpenAIRE graph
Found an issue? Give us feedback
addClaim

This Research product is the result of merged Research products in OpenAIRE.

You have already added 0 works in your ORCID record related to the merged Research product.

Защита Ð½ÐµÐ¹Ñ€Ð¾ÑÐµÑ‚ÐµÐ²Ñ‹Ñ ÑÐ¸ÑÑ‚ÐµÐ¼ компьютерного зрения от бэкдор-атак

выпускная квалификационная работа магистра

Защита Ð½ÐµÐ¹Ñ€Ð¾ÑÐµÑ‚ÐµÐ²Ñ‹Ñ ÑÐ¸ÑÑ‚ÐµÐ¼ компьютерного зрения от бэкдор-атак

Abstract

Цель работы – разработка прототипа нейросетевой системы компьютерного зрения с механизмом защиты от бэкдор-атак. Применено специализированное программно-математическое обеспечение pycharm. В результате исследования был проведен обзор систем компьютерного зрения и существующих методов атак на них, а также методов борьбы с этими атаками. В работе была предложена новая методика защиты, основанная на использовании хэш функций и протокола TLS для защиты канала передачи данных, которая показала свою эффективность в защите пользовательских данных от подмены. При разработке прототипа нейросетевой системы были учтены сформированные требования к системе и на их основе сформирована архитектура системы. Модель внедрения бэкдоров в изображения была разработана для демонстрации работы разработанного механизма выявления и смягчения бэкдоров. Механизм выявления и смягчения бэкдоров был разработан на основе анализа изменений в точности классификации данных при внедрении бэкдоров. Результаты работы алгоритма показали эффективность предложенной методики защиты от бэкдор-атак. Таким образом, результаты работы представляют новизну в области защиты нейросетевых систем компьютерного зрения от бэкдор-атак и могут быть использованы для улучшения безопасности подобных систем.

The object of research is neural network systems of computer vision. The purpose of the work is to develop a prototype of a neural network computer vision system with a mechanism to protect against backdoor attacks. Specialized software and mathematical software pycharm was used. As a result of the study, a review of computer vision systems and existing methods of attacks on them, as well as methods for combating these attacks, was carried out. The work proposed a new protection technique based on the use of hash functions and the TLS protocol to protect the data transmission channel, which has shown its effectiveness in protecting user data from spoofing. When developing a prototype of a neural network system, the formed requirements for the system were taken into account and, on their basis, the system architecture was formed. The image backdoor injection model was developed to demonstrate the operation of the developed mechanism for detecting and mitigating backdoors. The mechanism for detecting and mitigating backdoors was developed based on the analysis of changes in the accuracy of data classification when backdoors are introduced. The results of the algorithm showed the effectiveness of the proposed method of protection against backdoor attacks. Thus, the results of the work represent a novelty in the field of protecting neural network computer vision systems from backdoor attacks and can be used to improve the security of such systems.

Keywords

бэкдор, троянская атака, заражение данныÑ, компьютерное зрение, data infection, Нейронные сети, Информация, trojan attack, классификация изображений, backdoor, computer vision, image classification

  • BIP!
    Impact byBIP!
    citations
    This is an alternative to the "Influence" indicator, which also reflects the overall/total impact of an article in the research community at large, based on the underlying citation network (diachronically).
    0
    popularity
    This indicator reflects the "current" impact/attention (the "hype") of an article in the research community at large, based on the underlying citation network.
    Average
    influence
    This indicator reflects the overall/total impact of an article in the research community at large, based on the underlying citation network (diachronically).
    Average
    impulse
    This indicator reflects the initial momentum of an article directly after its publication, based on the underlying citation network.
    Average
Powered by OpenAIRE graph
Found an issue? Give us feedback
citations
This is an alternative to the "Influence" indicator, which also reflects the overall/total impact of an article in the research community at large, based on the underlying citation network (diachronically).
BIP!Citations provided by BIP!
popularity
This indicator reflects the "current" impact/attention (the "hype") of an article in the research community at large, based on the underlying citation network.
BIP!Popularity provided by BIP!
influence
This indicator reflects the overall/total impact of an article in the research community at large, based on the underlying citation network (diachronically).
BIP!Influence provided by BIP!
impulse
This indicator reflects the initial momentum of an article directly after its publication, based on the underlying citation network.
BIP!Impulse provided by BIP!
0
Average
Average
Average
Upload OA version
Are you the author? Do you have the OA version of this publication?