Knowledge Graph-Driven Electromagnetic Spectrum Map Completion for Spaceborne Sparse Observations

Abstract

Spaceborne observations are affected by revisit cycles, noise, and occlusion, resulting in measured data that are often spatially sparse and insufficient for high precision reconstruction of electromagnetic spectrum maps. This paper proposes a knowledge graph based spectrum map completion algorithm tailored for spaceborne sparse observation scenarios. The algorithm uses prior information stored in the knowledge graph—such as radiation source coordinates, frequencies, and transmit powers—to generate a theoretical power field via the free space propagation model. It then exploits the residuals between sparse observed powers and the theoretical values, and applies Gaussian smoothing to extend these residuals over the entire region, achieving a spatial fusion of theory and measurements. Experiments are conducted with unobserved point ratios ranging from 90.4% to 99%. Under the extreme sparsity of 99%, the algorithm achieves a coefficient of determination (R2) of 0.864 and a root mean square error (RMSE) of 1.74 dB for the unobserved points, both outperforming six comparison methods including K-nearest neighbors, Kriging, and inverse distance weighting. When the number of radiation sources increases from 3 to 9, the proposed method maintains stable accuracy, while the errors of the comparison methods generally increase. In summary, the algorithm enables high accuracy map reconstruction under sparse sampling, with strong robustness and good physical interpretability, providing an effective technical approach for spaceborne electromagnetic spectrum situational awareness.