Channel-Adaptive Diffusion Shield for Secure Image Semantic Communications

Author First name, Last name, Institution

• Taojie Zhu, Nanjing University of Post and TeleCommunications
• Qiang Wu, Ltd.
• Xinmian Xu, Nanjing University of Post and TeleCommunications
• Mingkai Chen, Nanjing University of Post and TeleCommunications
• Lei Wang, Nanjing University of Post and TeleCommunications
• Shahid Mumtaz, Nottingham Trent University
• Wael Bazzi, American University in Dubai
• Anwer Al-Dulaimi, Zayed University, Abu Dhabi CampusFollow

Document Type

Conference Proceeding

Source of Publication

2026 IEEE International Conference on Communications Workshops Icc Workshops 2026 Proceedings

Publication Date

5-24-2026

Abstract

In the era of the Industrial Internet of Things (IIoT), ensuring secure and reliable data transmission over harsh wireless channels is critical for resilient industrial networks. While semantic communication offers a spectral-efficient solution for sustainable industrial connectivity, it exposes sensitive operational data to eavesdropping risks. Existing secure semantic schemes often suffer from weak protection against semantic leakage and lack adaptability to dynamic industrial channel conditions. To address these challenges, we propose a Sustainable and Resilient Semantic Diffusion Shield (SuRe-SDS), which innovatively leverages channel noise - conventionally a detriment - as an effective medium for encryption. First, we design a noise-driven transmission encryption scheme that projects data into latent noise space, turning environmental interference into a security asset for low-overhead protection. Second, we introduce an asymmetric semantic diffusion module that fuses modified source semantics with encrypted noise, ensuring the transmitted signals bear no perceptual resemblance to the original privacy-sensitive imagery. Third, to enhance resilience against time-varying industrial channels, we develop a channel-adaptive diffusion step predictor. Extensive experiments demonstrate that SuRe-SDS significantly outperforms state-of-the-art methods in terms of semantic fidelity, robustness against channel impairments, and resistance to eavesdropping.

DOI Link

10.1109/iccworkshops63917.2026.11586640

ISBN

[9798331576240]

Publisher

IEEE

Disciplines

Computer Sciences

Keywords

diffusion models, eavesdropping prevention, noise-driven encryption, Resilient industrial networks, secure semantic communication

Scopus ID

105045579426

Recommended Citation

Zhu, Taojie; Wu, Qiang; Xu, Xinmian; Chen, Mingkai; Wang, Lei; Mumtaz, Shahid; Bazzi, Wael; and Al-Dulaimi, Anwer, "Channel-Adaptive Diffusion Shield for Secure Image Semantic Communications" (2026). All Works. 8189.
https://zuscholars.zu.ac.ae/works/8189

Indexed in Scopus

yes

Open Access

no