Channel Knowledge Map-Aided Semantic Image Compression for Lunar Surface Communications

Abstract

Lunar surface image transmission is essential for rover navigation, terrain interpretation, and scientific exploration. However, severe terrain blockage, spatially varying link conditions, and the high bandwidth demand of image data make fixed compression and transmission strategies inefficient in lunar surface environments. This paper proposes a CKM-aided link-aware semantic image transmission framework, which constructs a path-loss-oriented lunar surface channel knowledge map (CKM), converts the CKM-derived channel knowledge into local transmission capability, and uses it to guide the mode selection of a variable-rate Gained-VAE image compression model. A UNet-based CKM predictor is developed from sparse channel measurements, digital elevation model (DEM)-based terrain information, and environmental priors, achieving an RMSE of 0.0916 and an MAE of 0.0637 compared with ray-tracing-based reference CKMs. Trajectory-level transmission experiments are conducted in a 5 km by 5 km region near the Chang'E-4 landing area. Compared with fixed heavy compression, the proposed adaptive compression strategy maintains a high image transmission success rate across different predefined rover trajectories, while improving the average PSNR by up to 3.43 dB. This demonstrates that the proposed link-aware semantic image compression method achieves a better trade-off between transmission reliability and reconstruction quality over rover trajectories with complex channel conditions.