TA-UNet: A Satellite Electromagnetic Map Sparse Reconstruction Method With Past Sensing Information

Electromagnetic map reconstruction is a critical technology for communication satellites to be aware of the complete electromagnetic situation, serving as a fundamental basis for efficient spectrum sharing in satellite communications. The satellite electromagnetic environment exhibits significant short-term temporal correlation, with sampling points dynamically varying over time. Leveraging these characteristics, we propose a method for sparse reconstruction of satellite electromagnetic maps using past sensing information, and design a neural network, Temporal Attention UNet (TA-UNet). By incorporating a temporal attention mechanism, TA-UNet effectively integrates sensing information from multiple time slots, facilitating more efficient electromagnetic map reconstruction. Experiments demonstrate that past sensing information significantly improves the similarity between the reconstructed map and the ground truth. Moreover, TA-UNet demonstrates superior reconstruction accuracy and real-time performance, outperforming previous methods. Compared with UNet without past sensing information, the reconstruction error is reduced by 12.67% to 62.70%.