Few-Sample GNSS Environment Recognition via LLM-Based Physical-Semantic Distillation and Test-Time Adaptation
Global Navigation Satellite System (GNSS) environment recognition is important for enhancing positioning reliability and context awareness in complex urban and natural scenes. However, existing methods predominantly rely on signal features and necessitate extensive labeled datasets, compromising their robustness in label-scarce scenarios. To address this, we propose a large language model (LLM)-assisted, few-sample GNSS environment recognition framework. Our approach leverages LLM-generated physical-semantic soft labels for knowledge distillation during training and incorporates a lightweight test-time adaptation (TTA) strategy during inference. Experimental results demonstrate that with only 1% of labeled training data, the proposed method outperforms the baseline by 6% in classification accuracy.