Deep Learning-Based Defense Against GPS Spoofing Attacks in Unmanned Aerial Vehicles

Drones rely heavily on GPS sensors for navigation, making them vulnerable to GPS spoofing attacks that can manipulate flight paths and compromise operations. To address this critical challenge, we propose a novel detection method leveraging deep learning to identify spoofed GPS signals. Our detection model, BiLSTM-Attention-CNN, integrates a bidirectional long short-term memory (BiLSTM) layer to capture global temporal dependencies, an attention mechanism to focus on key sequence features, and a 1D convolutional neural network (1D-CNN) layer to extract local patterns effectively. We detail the development of a simulation environment using a modified version of JMAVSim to emulate realistic GPS spoofing scenarios. Using this setup, combined with QGroundControl for flight experiments, we systematically analyze various GPS spoofing techniques and collect comprehensive datasets for training and evaluation. Comparative experiments demonstrate the superiority of the BiLSTM-Attention-CNN model over both traditional machine learning and other deep learning approaches, achieving state-of-the-art performance in detecting GPS spoofing attacks. These findings underscore the potential of deep learning-based frameworks to enhance UAV navigation security.