High-Precision Detection of GPS Spoofing Attacks on UAVs Using MLP

This work addresses the problem of detecting GPS spoofing attacks on Unmanned Aerial Vehicles (UAVs) using a multilayer perceptron (MLP). Such attacks allow adversaries to inject artificial signals of increased power that confuse the drone’s navigation system and cause deviations from its planned route. The open-source TEXBAT dataset was used for experimental research, with separate highlights of the DS3 and DS7 scenarios that simulate synchronous GPS spoofing. During the data preparation stage, the signal parameters (pseudorange, power, and Doppler shift) were leveraged, and their statistical analysis was performed using correlation matrices and mean-value distributions. The proposed MLP model, featuring an optimized architecture with three hidden layers and a sigmoid activation function at the output, demonstrated a detection accuracy of 93% on the validation data. The advantages of this approach include high performance and ease of integration into navigation systems. However, the relatively small amount of real data limits scalability and comprehensiveness of the evaluation.