Semantic Adaptive Communication Based on Double-Attention Phase and Compress Estimator for Wireless Image Transmission

In existing semantic communication systems for image transmission, some images are generally reconstructed with considerably low quality and a high transmission rate. Driven by the imperative to effectively tackle these longstanding challenges, semantic communication has emerged as a critical technological advancement. In this work, we propose a Semantic Adaptive Communication (SAC) framework to transmit images with core information. Specifically, the proposed framework is composed of a Semantic Encoder (SE) and Semantic Decoder (SD), a Semantic Code Generator/Restore (SCG/SCR) module, a Compression Estimator (CE), a Channel State Information Acquisition (CSIA) module and a Wireless Channel. To fully capture both channel attention and spatial attention for semantic features, we design a Double-Attention Module (DAM) that operates alongside channel conditions, integrated into the SE and SD. Additionally, in order to predict the compress rate of the SAC, the CE works based on the channel condition and the recover quality. The experimental results demonstrate that the proposed SAC framework has a greater PSNR (increased by 0.5-2 dB) and accuracy value (91-93%), which indicate the SAC robustness, than traditional communication methods and other semantic communication algorithms in image transmission scenarios. In addition, the proposed framework achieves adaptive transmission rates with minimal sacrifice in recovery performance while enhancing the bandwidth utilization efficiency of the semantic communication system.