太赫兹科学与电子信息学报2025，Vol.23Issue(10)：1032-1037,6.DOI:10.11805/TKYDA2024131
面向鲁棒多级语义通信系统中的对抗攻击
Adversarial attacks in robust multi-level semantic communication systems
陈军 1史明佳 2林云2
作者信息
- 1. 成都富元辰科技有限公司,四川 成都 610037
- 2. 哈尔滨工程大学 信息与通信工程学院,黑龙江 哈尔滨 150006
- 折叠
摘要
Abstract
With the rapid advancement of modern information technologies,the number of Internet of-Things devices has exploded,urgently calling for new communication paradigms.Semantic communication systems have demonstrated great potential for boosting end-to-end transmission performance and represent a key research direction for future communications.However,deep-learning-based semantic schemes suffer from security vulnerabilities:minute perturbations can mislead the model and block real-world deployment.This paper proposes a deep-learning-driven multi-level semantic communication system that jointly transmits data to accomplish downstream tasks.Realistic channel impairments and adversarial attacks are incorporated to reflect practical scenarios.Experiments show that the proposed system exhibits a certain level of robustness,offering insights for designing more resilient semantic communication systems.关键词
语义通信/信源信道联合/对抗攻击/图像传输Key words
semantic communication/source-channel association/adversarial attacks/image transmission分类
电子信息工程引用本文复制引用
陈军,史明佳,林云..面向鲁棒多级语义通信系统中的对抗攻击[J].太赫兹科学与电子信息学报,2025,23(10):1032-1037,6.
