When Does Multimodal AI Help? Diagnostic Complementarity of Vision-Language Models and CNNs for Spectrum Management in Satellite-Terrestrial Networks

Researchers developed SpectrumQA, a benchmark comparing vision-language models (VLMs) and CNNs for spectrum management in satellite-terrestrial networks. The study reveals task-dependent complementarity: CNNs excel at spatial localization while VLMs uniquely enable semantic reasoning capabilities that CNNs lack entirely.

• →VLMs and CNNs show complementary strengths in spectrum management tasks rather than being direct substitutes.
• →CNNs achieved 72.9% accuracy in severity classification and 0.552 IoU in spatial localization tasks.
• →VLMs uniquely enabled semantic reasoning with F1=0.576 using only three examples, a capability absent in CNN architectures.
• →A hybrid approach using both models achieved 39.1% improvement over CNN-only solutions.
• →VLM representations showed stronger cross-scenario robustness compared to CNNs in transfer learning tasks.