Key points are not available for this paper at this time.
Drone communication networks – especially those relying on Drone-to-Drone (Dr2Dr) and Drone-to-Base Station (Dr2BS) links – are increasingly vulnerable to threats like DDoS, GPS spoofing, and jamming. Their high mobility, wireless architecture, and limited computational resources make traditional intrusion detection systems (IDS) less effective in securing these networks. To address this, we propose FL-DMAN, a Federated Learning-enabled Dynamic Multi-Scale Attention Network for real-time threat detection. FL-DMAN combines multi-scale feature extraction with a dynamic attention mechanism, accurately detecting complex attack patterns while maintaining low computational overhead. Using federated learning, the model trains across distributed drone nodes without sharing raw data, ensuring efficiency and privacy. An enhanced Giant Trevally Optimization (EGTO) algorithm is also integrated to fine-tune hyperparameters and improve convergence speed. FL-DMAN achieves 99.0% accuracy and 99.4% AUC on real-world and simulated datasets, with low false positive rates—even in high-traffic conditions. The framework is designed for scalability and supports large drone fleets and swarm-based applications, making it suitable for defence, disaster response, and autonomous delivery systems. Future work will optimize the model for edge deployment and explore next-generation security mechanisms for evolving drone ecosystems.
Banjar et al. (Mon,) studied this question.