Zero-Shot Learning-Based Maritime Semantic Communication System

Semantic communication (SC) is critical for the efficient and accurate transmission of data in complex environments, especially for unmanned surface vehicles (USVs) operating in marine settings characterized by strong interference and limited bandwidth. Traditional methods of transmitting large volumes of raw data are inefficient and susceptible to errors. Additionally, semantic communication systems, which depend on labeled datasets, have limited capacity to handle unfamiliar objects in dynamic marine environments. The long-tailed data distribution further complicates the recognition and processing of novel data types. To address these issues, this paper presents a novel approach that integrates zero-shot learning (ZSL) with SC, effectively enhancing image transmission and recognition tasks accuracy in various conditions. The proposed method employs semantic encoding and decoding to prioritize meaningful data, thereby enhancing transmission efficiency and accuracy. Additionally, we introduce semantic metrics for training and evaluating system performance, improving the ability to handle unfamiliar data types. Experimental results demonstrate that our algorithm significantly outperforms traditional methods in image transmission.