A Nested UNet-Based Maritime Image Semantic Communication System

The demand for image transmission in maritime communication is constantly increasing. However, the limited nature of spectrum resources restricts the transmission of massive amounts of image data. In this paper, we propose the nested UNet-based maritime image semantic communication system (NU-MSCS), built upon the UNet++ architecture. NU-MSCS replaces the conventional syntactic-level encoder with a multi-level semantic fusion encoder and integrates joint semantic channel coding (JSCC) to improve robustness under noisy conditions. Furthermore, a deep supervision (DS) mechanism is introduced to accelerate training convergence and enhance transmission reliability. Simulation results show that the proposed NU-MSCS has desirable properties in terms of communication reliability and training efficiency, providing a promising solution for image transmission in maritime communications.