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

Unmanned Aerial Vehicle swarm networks are increasingly deployed in critical missions such as surveillance, disaster management, and military coordination, where secure communication is paramount. However, their decentralized topology and dynamic mobility make them highly vulnerable to sophisticated cyber-attacks. To address these challenges, this research presents a novel, memory-efficient Residual-inspired One-Dimensional Convolutional Neural Network framework for high-fidelity intrusion detection in UAV swarm environments. The proposed architecture utilizes residual skip connections to enhance gradient flow, stabilize deep learning, and efficiently extract discriminative features from UAV network traffic. It further incorporates dynamic class-weighting to mitigate data imbalance and a comprehensive memory-time profiling mechanism to ensure deployability in resource-constrained aerial systems. Experimental evaluations on the publicly available UAVIDS-2025 dataset demonstrate outstanding performance, achieving an accuracy of 99.71%, a macro F1-score of 0.9971, and a ROC-AUC of 0.9999. The model maintains a compact parameter footprint of 474,629 trainable parameters (1.81 MB) and an inference time of approximately 0.0023 seconds per sample. These results affirm that the proposed framework delivers superior detection precision, computational efficiency, and scalability, positioning it as a robust and deployable solution for next-generation UAV network security and intelligent cyber-defense systems.