A Cross-Scenario GNSS Spoofing Detection Method Based on Transfer Learning

As a critical information infrastructure, the reliability of positioning, navigation, and timing (PNT) services from Global Navigation Satellite Systems (GNSS) is directly related to national security and economic operation. However, the open nature of satellite signals and significant propagation path loss make them susceptible to spoofing attacks. Existing spoofing detection methods exhibit poor cross-scenario generalization, causing performance degradation with scarce labeled data and high false alarm rates in multipath environments. To address these issues, we propose a cross-scenario detection method that employs a Transformer to capture long-term signal dynamics and a pre-training/fine-tuning transfer learning framework to ensure rapid adaptation to new data domains. Experimental results demonstrate that the proposed transfer learning framework improve detection accuracy by up to 10.54% and reduce training time by 75.6% using only 5% of labeled data from a new domain. Furthermore, the model demonstrates high robustness in complex scenarios, achieving a 99.75% detection rate against a 0.03% false alarm rate in concurrent spoofing and multipath environments. This work provides a highly robust solution with cross-scenario adaptability for anti-spoofing protection in complex electromagnetic environments.