EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

A Black-Box Hybrid-Attack for Image Semantic Communication Systems

Kunqiang Ai, Donghong Cai, Mohammad S. Obaidat, Fang Fang, Lisheng Fan

IEEE Open Journal of the Communications Society · 2026 · 7 · 4471-4482

DOI: 10.1109/ojcoms.2026.3688319 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Open Journal of the Communications Society, 2026. Verfasst von Kunqiang Ai und 4 weiteren Autor:innen. Die Arbeit nennt 42 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Basic and Applied Basic Research Foundation of Guangdong Province.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Semantic communication (SC) systems employ deep neural networks to transmit semantic features rather than raw data, thereby improving transmission efficiency and supporting next-generation wireless applications. This paper proposes a black-box hybrid attack that reveals critical vulnerabilities in image-based SC systems performing both reconstruction and classification tasks. Unlike existing black-box approaches that depend on surrogate models trained on datasets with distributions similar to the target, our method generates a synthetic substitute dataset, removing this dependency. The proposed attack consists of two stages. In the first stage, a full-duplex attacker intercepts semantic features transmitted over an open channel, reconstructs counterfeit images using a neural network, and simultaneously disrupts the initial communication. In the second stage, semantic noise is generated from these counterfeit images and injected into the retransmitted features, misleading the receiver while maintaining perceptual image quality. A dedicated data-generator network enhances the quality of fake data and strengthens the semantic noise design. Experiments conducted on CIFAR-10 and MNIST demonstrate that our method reduces classification accuracy by up to 54%, while preserving high visual fidelity with PSNR $\mathit {approx}~26$ dB and SSIM $\mathit {approx}~0.95$ . These findings highlight severe security risks in SC systems and emphasize the urgent need for more robust defense mechanisms.

Förderung

  • Basic and Applied Basic Research Foundation of Guangdong Province

Zitieren

APA

Ai, K., Cai, D., Obaidat, M. S., Fang, F., & Fan, L. (2026). A Black-Box Hybrid-Attack for Image Semantic Communication Systems. IEEE Open Journal of the Communications Society, 7, 4471-4482. https://doi.org/10.1109/ojcoms.2026.3688319

Vancouver

Ai K, Cai D, Obaidat MS, Fang F, Fan L. A Black-Box Hybrid-Attack for Image Semantic Communication Systems. IEEE Open Journal of the Communications Society. 2026;7:4471-4482. doi:10.1109/ojcoms.2026.3688319

BibTeX
@article{Ai2026,
  author = {Ai, Kunqiang and Cai, Donghong and Obaidat, Mohammad S. and Fang, Fang and Fan, Lisheng},
  title = {A Black-Box Hybrid-Attack for Image Semantic Communication Systems},
  journal = {IEEE Open Journal of the Communications Society},
  year = {2026},
  volume = {7},
  pages = {4471--4482},
  doi = {10.1109/ojcoms.2026.3688319}
}
Kostenlos starten →