华中科技大学学报（自然科学版）2024，Vol.52Issue(11)：1-7,7.DOI:10.13245/j.hust.241101
基于变分编码器与主成分分析的电磁对抗攻击
Electromagnetic adversarial attack based on variational auto-encoder and principal component analysis
摘要
Abstract
Aiming at the issue of the inability to promptly acquire distinct original input signals for electromagnetic signals,which consequently obstructed the generation of effective adversarial samples,a universal adversarial attack method based on variational auto-encoder(VAE)and principal component analysis(PCA)was proposed.The VAE and PCA based universal adversarial perturbation(VP-UAP)first sampled the input signal and generated an adversarial perturbation set using a traditional adversarial attack method.The VAE and PCA were then used for dual-feature extraction on the adversarial perturbation set,leveraging the extracted universal perturbation features to generate universal adversarial perturbations,and thereby accomplishing universal adversarial attacks in the black-box scenario.Experimental results on RML2016.10a show that VP-UAP exhibits great attack efficacy on different models.Furthermore,compared to other baseline algorithms,better adaptability and attack performance are exhibited when facing signals of different signal-to-noise ratios(SNR)and various perturbation-to-noise ratios(PNR).关键词
调制分类/深度识别模型/电磁领域/通用对抗攻击/黑盒攻击Key words
modulation classification/deep recognition model/electromagnetic field/universal adversarial attack/black box attack分类
信息技术与安全科学引用本文复制引用
何琨,陈振华,黄绍,赵耀东..基于变分编码器与主成分分析的电磁对抗攻击[J].华中科技大学学报（自然科学版）,2024,52(11):1-7,7.基金项目
国家自然科学基金资助项目(U22B2017) （U22B2017）
电磁空间安全全国重点实验室基金资助项目. （）
