LS-SSC: Lightweight Stealth-Aware Secure Semantic Communication Against Eavesdroppers

As an emerging communication paradigm, semantic communication (SC) has garnered increasing attention, but its open transmission characteristics and deep learning dependency pose severe security challenges. In this paper, we propose a lightweight stealth-aware secure SC (LS-SSC) framework with adaptive weighting to balance adversarial and stealthy training objectives for a resource-constrained image transmission syst em under eavesdropping threats. By introducing a channel-spatial attention-enhanced reversible residual network, LS-SSC can conduct stealthy adversarial attacks against eavesdroppers, stealthily disrupting the semantic information perceived by eaves-droppers. Experimental results show that compared with the baseline scheme, the proposed LS-SSC saves 71% parameters and 41% floating point operations. Moreover, LS-SSC can effectively attack the eavesdropper's downstream artificial intelligent task network while guaranteeing strong stealthiness for the attack behavior, and meanwhile it can improve the decoding quality for the legitimate receiver.