Hybrid Multi-Head Attention and Multi-Scale CNN for Drone Network Intrusion Detection

The rapid evolution of the Internet of Drones (IoD) has introduced numerous advantages across various sectors; however, it has also increased exposure to diverse cyber threats. Ensuring the security of drone communications and operations has become a critical challenge due to the dynamic, distributed, and wireless nature of IoD environments. Traditional intrusion detection systems often fall short in identifying sophisticated and evolving attack patterns, especially when handling large-scale and heterogeneous network data. To address these limitations, a Multi-Head Attention based Frivolous Multi-scale Convolutional Neural Network (Multi-FMCNN) is employed for effective intrusion detection. This architecture enables the extraction of both spatial and contextual features from complex traffic patterns, improving the model’s ability to detect both known and unknown threats. To further enhance performance, the Polar Fox Optimization Algorithm is utilized to enhance the network parameters, thereby maximizing detection accurateness. Experimental validation is conducted using benchmark datasets— NSL-KDD, CICIDS2017, and UNSW-NB15 covering a wide range of attack categories. The suggested system attains a remarkable accuracy of 99.8%, demonstrating its robustness and adaptability in real-time IoD intrusion detection scenarios. This approach offers a significant advancement toward intelligent and resilient security frameworks for modern drone-based networks.