Residual-aware lightweight deep learning framework for high-fidelity intrusion detection in UAV swarm networks

Area of Science:

• Cybersecurity
• Artificial Intelligence
• Aerospace Engineering

Background:

• Unmanned Aerial Vehicle (UAV) swarms are critical for missions but vulnerable to cyber-attacks due to their decentralized nature.
• Existing intrusion detection systems struggle with the unique challenges of UAV swarm networks, including dynamic mobility and resource constraints.

Purpose of the Study:

• To develop a novel, memory-efficient intrusion detection framework for UAV swarm networks.
• To enhance the security and reliability of communication within UAV swarms against sophisticated cyber threats.

Main Methods:

• A Residual-inspired One-Dimensional Convolutional Neural Network (1D-CNN) architecture was designed.
• The framework incorporates residual skip connections for improved feature extraction and dynamic class-weighting to handle data imbalance.
• Memory-time profiling was implemented for resource-constrained environments.

Main Results:

• The proposed model achieved high performance on the UAVIDS-2025 dataset, with 99.71% accuracy, 0.9971 macro F1-score, and 0.9999 ROC-AUC.
• The model is computationally efficient, with 1.81 MB size and 0.0023 seconds inference time per sample.
• Demonstrated superior detection precision and scalability for UAV network security.

Conclusions:

• The novel deep learning framework offers a robust and deployable solution for securing UAV swarm communications.
• The research contributes to advancing intelligent cyber-defense systems for next-generation aerial networks.
• The memory-efficient design makes it suitable for resource-limited UAV applications.