A black-box attack method for brain-computer interface system

Abstract

The invention discloses a black-box attack method for a brain-computer interface system. Firstly, EEG samples with more information and diversity are generated based on query synthesis and active learning as a substitute model training sample, and then the substitute model is trained based on the training sample, so that The surrogate model can better approximate the target model. Finally, by generating adversarial samples on the trained surrogate model, the adversarial samples are used to carry out black-box attacks on the EEG brain-computer interface system, so that it can be misclassified on the target model. Compared with the traditional black-box attack method based on Jacobian matrix, it has better attack effect, can achieve the same or better attack performance with fewer inquiries, and the generated adversarial samples have very little noise. There is almost no difference from the original EEG signal in the time domain and frequency domain, and it cannot be easily detected, which greatly improves the attack efficiency of black-box attacks in brain-computer interface systems.

Description

technical field [0001] The invention belongs to the security field of an EEG-based brain-computer interface system, and more specifically relates to a black-box attack method for a brain-computer interface system. Background technique [0002] The brain-computer interface system is a real-time communication system that connects the brain and external electronic devices. The system directly collects the physiological electrical signals generated by the human brain, and then converts them into commands that can control external electronic devices, thereby replacing human natural limbs. Or language organ, communicate with the outside world, control the external environment. EEG is by far the most widely used input signal for BCI systems due to its ease of use, relatively low cost, and minimal risk to the user. In the brain-computer interface system, the machine learning module is the most important module. With the development of deep learning in recent years, some deep learni...

AI Technical Summary

Problems solved by technology

[0006] Aiming at the defects of the prior art, the purpose of the present invention is to propose a black-box attack method of the brain-computer interface system, which aims to solve the attack efficiency caused by the need for a large number of inquiries to collect enough information when training the replacement model in the prior art lower question

Images