Deep Learning-Based GNSS Spoofing Attack Detection for Quadrotor UAV

Quadrotor unmanned aerial vehicles (UAVs) face significant security risks from global navigation satellite system (GNSS) spoofing attacks, as evidenced in modern warfare scenarios. Traditional residual-based detectors often fail due to the time-varying residual distribution inherent in UAVs' nonlinear dynamics. This paper proposes a novel detection framework for GNSS spoofing attacks by leveraging reliable physical measurements from onboard sensors. A deep neural network (DNN) is trained to establish a nonlinear mapping between UAV states-including Euler angles, angular velocities, total lift, and their temporal features-and state estimation residuals. The proposed detector identifies attacks by evaluating deviations between the DNN-predicted residuals and actual residuals. Simulations validate the framework's efficacy across diverse flight regimes, demonstrating its potential for real-time onboard deployment.