SACE-JSCC: Semantic Attention-guided Chaotic Encryption for Secure Wireless Image Transmission

With the deep integration of artificial intelligence and wireless communication, Semantic Communication (SemCom) based on Deep Joint Source-Channel Coding (Deep JSCC) has emerged as a key technology for efficient image transmission. However, the broadcast nature of wireless channels makes Deep JSCC systems inherently vulnerable to eavesdropping, as they transmit structured semantic features directly. Existing security measures often introduce significant computational overhead or lack adaptability to the transmitted content. To address this security-fidelity trade-off, this paper proposes a Semantic Attention-guided Chaotic Encryption Deep JSCC (SACE-JSCC) scheme. Our key innovation is a lightweight, hierarchical chaotic encryption module guided by a semantic attention mechanism. This module identifies the importance of different feature components and applies adaptive encryption, providing stronger protection to more critical semantic information. While a legitimate receiver with the key can perfectly reconstruct the image, an eavesdropper only recovers a noisy, semantically meaningless output. Experimental results on the CIFAR-10 dataset demonstrate that SACE-JSCC effectively secures the transmission against eavesdropping while maintaining high reconstruction quality, showing only negligible performance degradation compared to its unencrypted counterpart and achieving an excellent balance between security and fidelity.