Attention fusion source channel coding for wireless image transmission networks
Yu Sun, Chao Li · 2025
The joint source and channel coding (JSCC) scheme of wireless communication networks is a newly emerged communication paradigm that mainly exploits deep learning (DL) to realize source and channel coding. However, the existing JSCC-based network cannot fully use the limited bandwidth resources to capture effective information about the image, resulting in serious distortion in the details of the reconstructed image. In this paper, we proposed a novel JSCC scheme for wireless image transmission networks, named Attention Fusion Source Channel Coding (AFSCC). Our proposed approach fuses attention mechanisms from different dimensions to enhance image details. The AFSCC can achieve adaptive variable channel conditions and dynamically adjust resource assignment strategy according to transmission content, signal-to-noise ratio, and bandwidth ratio. The attention mechanism enables efficient utilization of bandwidth resources for image reconstruction. Results demonstrate the proposed method is more robust in diverse channels and acquires better performance in image details.