Secure Navigation Through GPS/INS Integration: Comparative Analysis of Supervised Deep Learning and Kalman Filtering for Precision and Spoofing Detection

The main objective of this study is to present a comparative analysis of navigation security methods utilizing Kalman filtering and supervised deep learning algorithms, specifically a Convolutional Neural Network (CNN), within intelligent systems based on the Global Positioning System (GPS) and the Inertial Navigation System (INS). The GPS is the most widely used GNSS (Global Navigation Satellite System), as it is historically the oldest and is known for its accuracy and reliability. However, GPS suffers from several threats affecting its precision and security. Factors such as Non-Line-of-Sight conditions, signal blockage, and even intentional attacks degrade its performance in denied environments. To overcome these issues, embedded systems are commonly used as complementary sources of navigation data. The inertial navigation system is often preferred, relying on three accelerometers and three gyroscopes. The GPS/INS fusion data can be performed through several approaches, including Kalman filtering and deep learning algorithms. In this study, we conducted an analytical comparison of these two fusion methods in terms of navigation accuracy, computation time, and resilience to interference. The results demonstrate that the CNN-based approach remains highly accurate, achieving a total RMSE (Root Mean Square Error) of less than 1 meter and showing strong resilience to interference with an F1-score of 84.2% in detecting GPS spoofing attacks. However, this approach is computationally demanding. On the other hand, the Kalman filter-based method exhibits low resource consumption and faster computation time. Still, its accuracy degrades significantly during GPS outages, and it cannot detect GPS spoofing attacks.