基于深度学习的图像语义通信系统

摘要：语义通信是一种新颖的通信方式，可通过传输数据的语义信息提高带宽效率。提出一种用于无线图像传输的系统。该系统基于深度学习技术开发并以端到端（E2E）的方式进行训练。利用深度学习实现语义特征的提取和重建，在发送端提取信源信息不同类型和不同形式的语义特征，在接收端融合各类型语义特征进行目标语义恢复。仿真结果表明，与基准模型相比，所提模型在信道环境恶劣的情况下，具有更好的重建精度。

关键词：端到端通信；多级语义通信；图像压缩；图像传输

Abstract: Semantic communications is a novel form of communication that improves bandwidth efficiency by transmitting semantic information about data. A system for wireless image transmission is introduced， which is developed based on deep learning techniques and trained by an end-to-end (E2E) approach. Deep learning is used to extract and reconstruct semantic features, extract different types and forms of semantic features of source information at the sending end, and fuse various types of semantic features at the receiving end for target semantic recovery. The simulation results show that compared with the benchmark model, the proposed model has better reconstruction accuracy under the bad channel environment.

Keywords: end-to-end communication; multi-level semantic communication; image transmission