Detection of GPS Spoofing Attacks in UAVs Based on Adversarial Machine Learning Model.

Area of Science:

• Cybersecurity
• Aerospace Engineering
• Artificial Intelligence

Background:

• Wireless communication and automation are transforming mobility, with autonomous vehicles and unmanned aerial vehicles (UAVs) becoming prevalent.
• Global Positioning System (GPS) signals used for UAV navigation are vulnerable to cyberattacks, particularly GPS spoofing, due to unencrypted transmissions.
• Machine learning (ML) enhances UAV intrusion detection systems (IDSs), but adversarial machine learning (AML) poses new threats by exploiting ML models.

Purpose of the Study:

• To develop and present a novel UAV-IDS that leverages AML to improve the detection and classification of GPS spoofing attacks.
• To enhance the robustness and security of UAV systems against sophisticated cyber threats.
• To introduce a detection model focused on adversarial training and deep learning for physical layer security.

Main Methods:

• Implementation of an adversarial machine learning (AML) methodology within a UAV intrusion detection system (IDS).
• Development of a novel detection model utilizing adversarial training defense mechanisms.
• Application of advanced deep learning techniques for enhanced GPS spoofing attack detection and classification.

Main Results:

• The developed AML detection model achieved a high detection accuracy of 98% for GPS spoofing attacks.
• Demonstrated effectiveness in handling large-scale datasets and complex detection tasks.
• Validated the model's capability to significantly improve UAV system security and robustness.

Conclusions:

• Adversarial machine learning is crucial for enhancing the security of UAV intrusion detection systems against GPS spoofing.
• The proposed AML-based IDS offers a robust solution for protecting UAVs from cyber threats.
• Physical layer security, combined with advanced deep learning and adversarial training, is vital for future UAV security architectures.