The study introduces a novel adversarial attack method, TFAttack, that targets both time-domain and frequency-domain EEG signals using wavelet transform. This approach aims to balance attack performance and imperceptibility by leveraging the strengths of both domains. Extensive experiments demonstrate the effectiveness of TFAttack on three datasets and three deep learning models. The perturbations generated by TFAttack are barely perceptible to the human visual system, enhancing security in brainprint recognition systems. The study highlights the vulnerabilities of deep-learning models in brainprint recognition and emphasizes the need for robustness research in this field.
他の言語に翻訳
原文コンテンツから
arxiv.org
深掘り質問