Semantic Information Defense and Attack Detection for Consumer Semantic Communication

Semantic communication, recognized as a core paradigm for 6G intelligent systems, provides low-latency and high-efficiency data transmission by the ability to extract semantic features and generation capability of neural networks. However, this advantage also introduces severe challenges in semantic security, particularly due to the inherent vulnerability of neural networks to semantic attacks, which can severely compromise downstream intelligent tasks. To address this issue, we propose an innovative semantic information security framework. Specifically, we introduce a semantic denoising assistant consumer semantic security scheme, which performs semantic-level recovery, thereby significantly improving downstream task accuracy while maintaining system efficiency. Furthermore, to reduce the latency and computational overhead introduced by the semantic denoiser, a lightweight attack detection assistant consumer adaptive semantic security scheme is designed, enabling the receiver to detect the presence of semantic attacks in the channel and adaptively activate the denoiser. The proposed framework restores task-relevant semantic information through semantic consistency constraints rather than pixel-consistency reconstruction, offering adaptive, lightweight, and plug-and-play security for semantic communication systems. Experimental results validate the robustness and generalization capability of the framework under various channel conditions and attack types, providing critical support for the secure deployment of semantic communication in practical wireless environments. Specifically, the 20% improvement in classification accuracy after denoising and the sustained 98.5% attack detection rate. Compared to other baseline models, our semantic denoiser achieves a significantly greater improvement in classification accuracy.