회전익 항공기 기종 분류를 위한 실시간 음향 기반 경량 딥러닝 모델 비교연구

허준석¹ ; 김두영²^{, *}

1해군사관학교 인공지능학과 사관생도
2해군 중령(진)/해군사관학교 인공지능학과 부교수

Comparative Study of Lightweight Deep Learning Models for Real-time Acoustic Classification of Rotary-wing Aircraft

Junseok Heor¹ ; Dooyoung Kim²^{, *}

1Midshipman, Dept. of Artificial Intelligence, Republic of Korea Naval Academy
2CDR(Sel.), ROK Navy/Associate professor, Dept. of Artificial Intelligence, Republic of Korea Naval Academy

Correspondence to: ^*Dooyoung Kim Dept. of Artificial Intelligence, Republic of Korea Naval Academy 1, Jungwon-ro, Jinhae-gu, Changwon-si, Gyungsangnam-do, 51704, Republic of Korea Tel: +82-2-907-5246 E-mail: dykim07@navy.ac.kr

초록

본 연구는 회전익 항공기 음향 신호를 활용한 실시간 기종 분류를 위해 경량화된 딥러닝 구조를 제안한다. 입력 음향은 1초 단위로 분할 후 MFCC로 변환되어 시간-주파수 특징을 효과적으로 표현한다. CNN과 SqueezeNet을 기반으로 한 특징 추출기와 GRU 및 TCN 시계열 처리기를 조합한 다양한 구조를 비교한 결과, SqueezeNet-TCN 모델은 가장 적은 파라미터와 3 MB 미만의 모델 크기로 90 % 이상의 정확도를 달성하였다

Abstract

This study proposes a lightweight deep learning framework for real-time classification of rotary-wing aircraft using acoustic signals. The audio input is segmented into 1-second frames and converted into MFCC to effectively capture time-frequency features. By comparing architectures combining CNN or SqueezeNet feature extractors with GRU and TCN temporal processors, the SqueezeNet-TCN model achieved over 90 % classification accuracy with a model size under 3 MB.

Keywords:

Mel-frequency Cepstral Coefficient, Convolution Neural Networks, Rotary-wing Aircraft, Real-time Classification, Lightweight Deep Learning

키워드:

MFCC, 합성곱 신경망, 회전익 항공기, 실시간분류, 경량 딥러닝

References

Mi‑8 Helicopter, Global Aviation, https://globalflyaviation.com/mi‑8‑helicopter
CH-53 Technical Data, Rotorcraft.info, https://rotorcraft.info/fe/acft/1217
Reo Olson, Daniel H. Cress, “Passive Acoustic Range Estimation of Helicopters,” Technical Report EL-92-13, U.S. Army Corps of Engineers, 1992.
Cooling High-Power Radar Systems: A Thermal Technology Guide, Military Embedded Systems, https://militaryembedded.com/radar-ew/thermal-management/cooling-high-power-radar-systems-a-thermal-technology-guide, (검색일: 2022. 2. 11.)
김기태, 임요준, “ISAR 영상 기반 해상표적 식별을 위한 인공지능 연구,” 산업시스템학회지, Vol. 45, No. 2, pp. 12–19, 2022.
박준형, 박준성, “저불확실성 마이크로 도플러 영상과 초경량 합성곱 신경망을 이용한 소형 드론 분류,” IEEE Transactions on Image Processing. [https://doi.org/10.1109/TIP.2024.3388895]
김태완, 문호석, 김종환, “합성곱 신경망 기반의 헬기 기종 분류 모형 연구,” 제어로봇시스템학회 논문지, Vol. 26, No. 6, pp. 479–486, 2020.
Zhong, Jilong, Fan, Aigen, Fan, Kuangang, Pan, Wenjie, and Zeng, Lu, “Research on the UAV Sound Recognition Method Based on Frequency Band Feature Extraction,” Drones, Vol. 7, No. 6, 2023.
Wijnker, Dirk, van Dijk, Tom, Snellen, Mirjam, de Croon, Guido, and De Wagter, Christophe, “Hear-and-avoid for Unmanned Air Vehicles Using Convolutional Neural Networks,” International Journal of Micro Air Vehicles, Vol. 13, pp. 1–15, 2021. [https://doi.org/10.1177/1756829321992137]
Bai, Shaojie, Kolter, J. Zico, and Koltun, Vladlen, “An Empirical Evaluation of Generic Convolutional and Recurrent Networks for Sequence Modeling,” arXiv preprint, arXiv:1603.03947, , 2016.
Davis, S. B. and Mermelstein, P., “Comparison of Parametric Representations for Monosyllabic Word Recognition in Continuously Spoken Sentences,” IEEE Transactions on Acoustics, Speech, and Signal Processing, Vol. 28, No. 4, pp. 357–366, 1980. [https://doi.org/10.1109/TASSP.1980.1163420]
MI-8 Helicopter – Technical Overview, Rotorcraft.info, https://rotorcraft.info/fe/acft/1224
UH-1 Iroquois – Technical Overview, Rotorcraft.info, https://rotorcraft.info/fe/acft/842
이상현, 이성일, “딥러닝 기반 무인항공기 탐지를 위한 음향 데이터셋 구축에 관한 연구,” 디지털콘텐츠학회 논문지, Vol. 25, No. 2, pp. 327–334, 2024.
Ko, Nuri, Park, Jung-Ju, and Kim, Young-Hwan, “Helicopter Sound Recognition Using Mel-frequency Cepstral Coefficients and Hidden Markov Models,” IEEE Transactions on Consumer Electronics, Vol. 59, No. 1, pp. 197–202, 2013.
Iandola, Forrest N., Han, Song, Moskewicz, Matthew W., Ashraf, Khalid, Dally, William J., and Keutzer, Kurt, “SqueezeNet: AlexNet-level Accuracy with 50x Fewer Parameters and <0.5MB Model Size,” arXiv preprint, arXiv:1602.07360, , 2016.
Gandhimathi Alias Usha, S., Maheswari, P. Uma, Priya, M. Jeya, Jeevalakshmi, V. et al., “Lightweight Deep Learning Model for Wireless Capsule Endoscopy Image Classification Using SqueezeNet,” Proceedings of the 2025 International Conference on Visual Analytics and Data Visualization (ICVADV), 2025. [https://doi.org/10.1109/ICVADV63329.2025.10961830]
김현종, 길명선, 문양세, “DGA 도메인 탐지를 위한 LSTM과 GRU 모델 비교,” 2021 한국컴퓨터종합학술대회 논문집, 한국정보과학회, pp. 122–124, 2021. 6.
Bai, Shaojie, Kolter, J. Zico, and Koltun, Vladlen, “An Empirical Evaluation of Generic Convolutional and Recurrent Networks for Sequence Modeling,” arXiv preprint, arXiv:1803.01271, , 2018.
Donahue, Jeff, Hendricks, Lisa Anne, Guadarrama, Sergio, Rohrbach, Marcus, Venugopalan, Subhashini, Saenko, Kate, and Darrell, Trevor, “Long-term Recurrent Convolutional Networks for Visual Recognition and Description,” Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pp. 2625–2634, 2015. [https://doi.org/10.1109/CVPR.2015.7298878]
TensorFlow, “tf.keras.losses. SparseCategoricalCrossentropy,” https://www.tensorflow.org/api_docs/python/tf/keras/losses/SparseCategoricalCrossentropy
Kingma, Diederik P., and Ba, Jimmy, “Adam: A Method for Stochastic Optimization,” Proceedings of the 3rd International Conference on Learning Representations (ICLR), 2015.
전술통신체계 – C4I/통신, 한화시스템, https://www.hanwhasystems.com/kr/business/defense/c5i/communication01.do