| Title |
Automated Drone Classification For Internet Of Drones Based On A Hybrid Transformer Model |
| ID_Doc |
11194 |
| Authors |
Han B.; Yan X.; Su Z.; Hao J. |
| Year |
2022 |
| Published |
Proceedings - International Conference on Tools with Artificial Intelligence, ICTAI, 2022-October |
| DOI |
http://dx.doi.org/10.1109/ICTAI56018.2022.00125 |
| Abstract |
The growing number of drone serving in Smart Cities make it increasingly necessary to develop new technologies to assist traffic management in low-altitude airspace. The Internet of Drones (IoD) is a layered network providing efficient and safe airways for various drone inter-operable services. Automated Drone Classification (ADC) is a prerequisite in the IoD system that not only monitors flights and avoids collisions between drones but also detects malicious drone flights. In this work, we propose a Dual-stream Convolutional Transformer (DCT) model to carry out ADC tasks based on drone remote control signals. The DCT combines the self-attention mechanism and convolution to capture long-range dependencies of different frequencies/time steps and local invariant features. We evaluated the proposed method on two public datasets and compared the results with the baseline studies. The DCT model outperforms other baselines with the improvement of 3 % accuracy in multiclass classification and reaches 94.49 % and 98.73 % on two different datasets at 30 dB signal-to-noise ratio. Furthermore, the experiments also show the great potential of DCT model on out-of-distribution by the improvement of 7 % accuracy. The above results demonstrate that the proposed method is a promising solution for supporting the ADC task in reliable, fast, and inexpensive ways. © 2022 IEEE. |
| Author Keywords |
Automated Drone Classifi-cation; Deep Learning; Drone Remote Control Signals; Internet of Drones; Time-Frequency Analysis |
