Malicious ADS-B data Generation Based on Improved GAN
Jin Sheng Lei, Ruifang Jiang, Zhijun Wu · 2023
Automatic dependent surveillance broadcast (ADS-B) system sends messages over unencrypted wireless channels without any information integrity protection measures, and its messages are at risk of interception and tampering, which can easily lead to impersonation and forgery attacks. At present, although the ADS-B data anomaly detection model based on machine learning has excellent performance in predicting normal samples, the machine learning model may face different degrees of risk in each stage of its life cycle due to the existence of a large number of attackers in real scenes. To build secure and reliable machine learning systems, exploit potential vulnerabilities. Aiming at the ADS-B abnormal data detection model based on machine learning, this paper studies a construction method of poisoning data with strong applicability and establishes attack model. By injecting malicious data generated by the Generative adversarial network into the machine learning model, the performance of the trained model deteriorates and data misclassification occurs. Experimental results show that the malicious ADS-B data generation method proposed in this paper achieves good results, which lays a foundation for optimizing system defense technology and guaranteeing ADS-B security.