Adversarial Machine Learning for Robust and Secure UAV Detection in Consumer Applications
Ikram Ud Din, Ahmad Almogren, Joel J. P. C. Rodrigues · IEEE Transactions on Consumer Electronics · 2025
This research evaluates a cognitive AI model for unmanned aerial vehicles (UAV) detection using adversarial machine learning (AML) techniques. We test the model using the VisDrone dataset across various environments, identifying significant performance variations, notably an accuracy drop from 95% to 87% in urban nighttime scenarios. To counter adversarial threats, we integrate AML strategies including adversarial training, detection filtering, and defensive distillation. Advanced configurations improved detection accuracy and precision to approximately 90%, highlighting the effectiveness of these techniques. Our study demonstrates the importance of robust and secure AI-driven UAV detection systems, paving the way for future real-world applications and resilience assessments against adversarial attacks.