Design of Semantic Communication System based on Multi-level Semantic Feature Extraction

Semantic communication focuses on the transmission of meaning. Existing semantic communication systems mostly rely on single-level semantic information, which limits their performance in different scenarios. In this paper, we propose a semantic communication system based on multi-level semantic feature extraction (MSF-SCS). It utilizes a dynamic semantic knowledge base to extract and integrate semantic information from different levels, including visual expressions, semantic descriptions, and structured tags. They form a knowledge network to assist the encoding and decoding process of semantic communication. In the communication backbone network, the proposed MSF-SCS incorporates a feedback mechanism based on channel characteristics and performance metrics. This enables dynamic updates to the encoding and decoding networks while ensuring that semantic information is fully expressed. Simulation results demonstrate that the proposed MSF-SCS can adapt to different channel environments and achieve better transmission performance, exhibiting desirable properties across various channel conditions. This approach offers a promising solution for the application of semantic communication.