Resilient UAV network security: Cross-attentive alignment of sparse and temporal features for adversarial intrusion detection

Ali G., Awang M. K., Rashid J., Hamza M., YAHYA H., Ghani A.

Vehicular Communications, cilt.62, 2026 (SCI-Expanded, Scopus)

  • Yayın Türü: Makale / Tam Makale
  • Cilt numarası: 62
  • Basım Tarihi: 2026
  • Doi Numarası: 10.1016/j.vehcom.2026.101094
  • Dergi Adı: Vehicular Communications
  • Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus, INSPEC
  • Anahtar Kelimeler: Adversarial machine learning (FGSM), Edge computing, Feature fusion, Intrusion detection system (IDS), Multi-head cross-attention, TabNet, Transformer encoder, UAV network security
  • İstanbul Gelişim Üniversitesi Adresli: Evet

Özet

The rapid adoption of unmanned aerial vehicle (UAV) communication networks in infrastructure has expanded the attack surface for complex cyberattacks, such as polymorphic intrusions and adversarial evasions. Standard deep learning (DL) models have limitations in simultaneously capturing the sparse, attribute-rich nature of network logs and the sequential dependencies of dynamic traffic flow. To address this gap, this article proposes a novel adversarial hybrid framework that integrates TabNet and a Transformer Encoder via a Multi-Head Cross-Attention (MHCA) mechanism. MHCA is more flexible than naive concatenation; it dynamically aligns TabNet's interpretable feature selection with the Transformer's temporal features to create a holistic view of threat vectors. In addition, the proposed framework directly incorporates the Fast Gradient Sign Method (FGSM) to optimize adversarial training and learn imperceptible perturbations. The proposed model is evaluated on the CIC-IDS2017 and UNSW-NB15 datasets, achieving accuracies of 99.47% and 98.24%, respectively. Most importantly, the adversarially trained model is robust and improves recall by 33.07% against minority Botnet attacks compared with standalone baseline models, while maintaining a small footprint (0.37-1.24 MB) suitable for resource-constrained edge deployment. This paper sets a precedent for safe, robust, and explainable UAV communication networks.