Abstract: The low altitude electromagnetic environment is becoming increasingly complex, and communication frequency bands are often subject to various interferences, resulting in abnormal or even widespread interruption of spectrum data. It is difficult to accurately provide a basis for dynamic adjustment of communication frequency bands and cannot guaran-tee the reliable operation of unmanned aerial vehicle collaborative tasks. However, existing reconstruction methods mostly rely on pure numerical driving, only filling in based on local correlations of data, lacking semantic understand-ing of the causes of anomalies and high-level contextual modeling, making it difficult to achieve accurate adaptive reconstruction in dynamic environments. This paper proposes a spectrum reconstruction method based on text se-mantics to address this issue. This method enhances the reconstruction problem from passive filling to active infer-ence by introducing textual descriptions to help the model understand why data is lost. Specifically, firstly, a mapping mechanism is constructed from anomalous semantic descriptions to spectral features, injecting human experience and physical laws into the model in textual form; At the same time, probability modeling of data distribution is achieved through diffusion models to achieve spectrum reconstruction. The experiment shows that this method significantly improves the reconstruction accuracy of spectrum data in dynamic scenarios, providing a more robust technical guar-antee for unmanned aerial vehicle spectrum monitoring tasks.

Key words: Spectrum reconstruction, text knowledge, large language model, diffusion model