Table 1 Comparison of Existing GPS Spoofing Detection Methods.
Ref. | Approach | Key parameters | Limitations/Strengths |
|---|---|---|---|
DeepPOSE (CNN + RNN) for trajectory correction | Sensor data (accelerometer, gyroscope, GPS) | High computational cost, real-time performance not tested | |
PERDET (ML-based UAV spoofing detection) | Perception data (accelerometer, gyroscope, magnetometer, GPS, barometer) | Limited dataset size, generalization not fully validated | |
Dynamic classifier selection for GPS spoofing | 10 ML classifiers, feature selection | Increased processing time due to dynamic selection | |
1D CNN-based anti-spoofing model | GPS signal characteristics | Limited comparison with traditional ML models | |
MLP-based GPS spoofing detection | Statistical features from base station path loss | Accuracy drops significantly with fewer base stations | |
Deep learning-based ATC spoofing detection | ADS-B signals, zero-bias DNN | Requires continuous learning for adaptation | |
ML-based cross-technology spoofing detection | Physical-layer details of ZigBee | Limited to specific cross-technology attacks | |
MLP-based UAV GPS spoofing detection | Flight data and GPS signals | Accuracy varies across datasets (TEXBAT: 83.23%, MAVLINK: 99.93%) | |
LTME: Cryptography + Trust Management for UAVs | Secure message encryption and authentication | Focuses on message security rather than spoofing detection | |
Reinforcement learning-based clustering for UAVs | Clustering strategy, network topology | Not specifically designed for GPS spoofing detection | |
This Study | Lightweight Deep Learning Model for UAV GPS Spoofing Detection | Fine Tuning and Reducing Model Size | Improves generalization, enhances detection accuracy |
