Page 62 - ITU Journal Future and evolving technologies Volume 2 (2021), Issue 5 – Internet of Everything

ITU Journal on Future and Evolving Technologies, Volume 2 (2021), Issue 5

based on the system requirements, i.e., the desired clas‑ [2] Dimas Pristovani Riananda, Galih Nugraha, Har‑
si ication performance and false alarm rate. We recog‑ ish Mahatma Putra, Muhammad Lukman Baid‑
nize that a complete consideration of out‑of‑library clas‑ howi, and Riza Alaudin Syah. “Smart pulley work‑
si ication requires adding out‑of‑library data in the train‑ low in delivery drone for goods transportation”.
ing set or adaptation of open set recognition by introduc‑ In: AIP Conference Proceedings. Vol. 2226. 1. 2020,
ing an OpenMax layer, which estimates the probability of p. 060010.
an input being from an unknown class [38]. On the other [3] Connie A Lin, Karishma Shah, Lt Col Cherie Maun‑
hand, our proposed model gives very low model uncer‑ tel, and Sachin A Shah. “Drone delivery of medica‑
tainty for in‑library signals, and this therefore still pro‑
tions: Review of the landscape and legal consider‑
vides a reasonable solution described above to detect the
ations”. In: The Bulletin of the American Society of
out‑of‑library drones in practice.
Hospital Pharmacists 75.3 (2018), pp. 153–158.
[4] Vinay Chamola, Vikas Hassija, Vatsal Gupta, and
6. CONCLUSION
Mohsen Guizani. “A Comprehensive Review of the
In this study, we proposed a system that uses drone con‑ COVID‑19 Pandemic and the Role of IoT, Drones,
troller RF signals to classify drones of different makes and AI, Blockchain, and 5G in Managing its Impact”. In:
models for a wide variety of SNRs. We used CNN classi‑ IEEE Access 8 (2020), pp. 90225–90265.
iers with two different sources for training the models: [5] Walid Saad, Mehdi Bennis, and Mingzhe Chen.
time‑series images, and spectrogram images. We showed “A Vision of 6G Wireless Systems: Applications,
that the CNN model using the spectrogram images is more Trends, Technologies, and Open Research Prob‑
resilient to noise when compared with the time‑series lems”. In: IEEE Network 34.3 (2020), pp. 134–142.
images based model. The proposed method that uses a DOI: 10.1109/MNET.001.1900287.
merged training set of RF signals at different SNR lev‑
els along with the proposed denoising mechanism was [6] M. Ritchie, F. Fioranelli, H. Grif iths, and B. Torvik.
shown to be effective for UAV classi ication even at SNRs “Micro‑drone RCS analysis”. In: Proc. IEEE Radar
not directly considered by the trained model. We also ex‑ Conf. Johannesburg, South Africa, Oct. 2015,
plored classi ication performance against training set size pp. 452–456.
and showed that reasonable classi ication accuracy can [7] Ismail Guvenc, Farshad Koohifar, Simran Singh,
still be obtained with limited training data. Consequently, Mihail L Sichitiu, and David Matolak. “Detection,
adding new classes to the model (e.g., to include data from tracking, and interdiction for amateur drones”. In:
newly released drones) does not entail a high computa‑ IEEE Commun. Mag. 56.4 (2018), pp. 75–81.
tion cost. Finally, we examined the model behavior with
[8] Crown Consulting. NASA Urban Air Mobility (UAM)
in‑library and out‑of‑library drone signals and concluded
Market Study. URL: https : / / ntrs . nasa . gov /
that the proposed model shows a good performance iden‑
citations/20190026762.
tifying drones from an unknown class. Our future work
[9] Federal Aviation Agency. Proposed Rule on Remote
includes comparing our results with federated learning
techniques, and testing of the proposed CNN‑based UAV Identi ication of Unmanned Aircraft Systems.
classi ication technique at a larger scale, such as using the [10] H. Zhang, C. Cao, L. Xu, and T. A. Gulliver. “A UAV
AERPAW experimental platform at NC State University. Detection Algorithm Based on an Arti icial Neural
Network”. In: IEEE Access 6 (May 2018), pp. 24720–
ACKNOWLEDGMENT 24728.

This work has been supported in part by NASA under [11] W. Zhou, L. Wang, B. Lu, N. Jin, L. Guo, J. Liu, H. Sun,
the Federal Award ID number NNX17AJ94A. The authors and H. Liu. “Unmanned Aerial Vehicle Detection
would like to thank Martins Ezuma at NC State University Based on Channel State Information”. In: Proc. IEEE
for providing the drone controller RF data set used in this Int. Conf. Sensing Commun. Netw. (SECON). Hong
study. Kong, China, June 2018, pp. 1–5.
[12] M. Z. Anwar, Z. Kaleem, and A. Jamalipour. “Ma‑
REFERENCES chine Learning Inspired Sound‑Based Amateur
Drone Detection for Public Safety Applications”.
[1] Hazim Shakhatreh, Ahmad H Sawalmeh, Ala Al‑ In: IEEE Trans. Veh. Technol. 68.3 (Jan. 2019),
Fuqaha, Zuochao Dou, Eyad Almaita, Issa Khalil, pp. 2526–2534.
Noor Shamsiah Othman, Abdallah Khreishah,
and Mohsen Guizani. “Unmanned aerial vehicles [13] Y. Seo, B. Jang, and S. Im. “Drone Detection Us‑
ing Convolutional Neural Networks with Acoustic
(UAVs): A survey on civil applications and key
STFT Features”. In: Proc. IEEE Int. Conf. Advanced
research challenges”. In: IEEE Access 7 (2019),
Video Signal Based Surveillance (AVSS). Nov. 2018,
pp. 48572–48634.
pp. 1–6.

50 © International Telecommunication Union, 2021