A study on the weakly-supervised deep learning algorithm for active sonar target recognition based on pseudo labeling using convolutional recurrent neural network model

Yena You; Wonnyoung Lee; Seokjin Lee

doi:10.7776/ASK.2024.43.5.502

All Issue

2024 Vol.43, Issue 5 Preview Page Next Page

Research Article

A study on the weakly-supervised deep learning algorithm for active sonar target recognition based on pseudo labeling using convolutional recurrent neural network model 합성곱 순환 신경망 모델을 이용한 의사 레이블링 기법 기반 능동소나 표적 식별 약지도 딥러닝 알고리즘 연구

30 September 2024. pp. 502-510

PDF XML

Abstract

In this paper, we proposed the weakly-supervised deep learning algorithm for active sonar target recognition based on pseudo labeling using Conventional Recurrent Neural Network (CRNN) model widely used for acoustic signal processing because it can effectively utilize small and unbalanced active sonar data. Active sonar simulation data assuming two different SNRs and clutter environments were used in the training and testing process, and spectrogram obtained by applying Short Time Fourier Transform (STFT) to the simulation data was used as a feature factor for algorithm training. The algorithm proposed in this paper was evaluated based on the target and nontarget F1-score using test data independent of training data. As a result, it was confirmed that the CRNN model showed significant performance not only in typical acoustic signal processing but also active sonar target recognition. Also, pseudo-labeling helps to improve the performance of the active sonar target recognition algorithm used the CRNN model.

Keywords

Active sonar

Target recognition

Weakly-supervised learning

Conventional Recurrent Neural Network (CRNN)

Pseudo labeling

본 논문은 음향 신호처리에 널리 사용되는 합성곱 순환 신경망(Convolutional Recurrent Neural Network, CRNN) 모델을 기반으로 의사 레이블링 기법을 적용하여 소량 및 불균형 능동소나 데이터를 효과적으로 활용할 수 있는 능동소나 표적 식별을 위한 약지도 딥러닝 알고리즘을 제안한다. 두 가지의 서로 다른 신호대잡음비와 클러터 환경을 가정하여 생성한 능동소나 시뮬레이션 데이터를 학습 및 테스트 과정에 사용하였으며, 시뮬레이션 데이터에 단시간 푸리에 변환(Short Time Fourier Transform, STFT)을 적용하여 얻은 스펙트로그램을 알고리즘 학습을 위한 특징 인자로 사용하였다. 본 논문에서 제안하는 알고리즘은 학습 데이터와 무관한 테스트 데이터를 사용하여 표적과 비표적 F1 점수를 지표로 성능을 평가하였으며, 그 결과 합성곱 순환 신경망 모델이 일반적인 음향 신호 처리뿐만 아니라 능동소나 표적 식별에서도 유의미한 성능을 보이는 것을 확인하였다. 또한 의사 레이블링 기법이 합성곱 순환 신경망 모델을 이용한 능동소나 표적 식별 알고리즘의 성능 개선에 도움을 주는 것을 확인할 수 있었다.

키워드

능동소나

표적 식별

약지도 학습

합성곱 순환 신경망

의사 레이블링

References

A. D. Waite, SONAR for Practicing Engineers,3rd ed. (Wiley, Hoboken, 2001), introduction.

G. De Magistris, P. Stinco, J. R. Bates, J. M. Topple, G. Canepa, G. Ferri, A. Tesei, and K. L. Page, "Automatic object classification for low-frequency active sonar using convolutional neural networks," Proc. OCEANS, 1-6 (2019).

10.23919/OCEANS40490.2019.8962860

D. P. Williams, "Underwater target classification in synthetic aperture sonar imagery using deep convolutional neural networks," Proc. 23rd ICPR, 2497-2502 (2016).

10.1109/ICPR.2016.7900011

F. Liu, T. Shen, Z. Luo, D. Zhao, and S. Guo, "Underwater target recognition using convolutional recurrent neural networks with 3-D Mel-spectrogram and data augmentation," Appl. Acoust. 178, 107989 (2021).

10.1016/j.apacoust.2021.107989

B. Shi, X. Bai, and C. Yao, "An end-to-end trainable neural network for image-based sequence recognition and its application to scene text recognition," IEEE Trans. Pattern Anal. Mach. Intell. 2298-2304 (2016).

10.1109/TPAMI.2016.2646371

X. Zheng, H. Chen, and Y. Song, "Zheng ustc team's submission for dcase2021 task4-semi-supervised sound event detection," DCASE, Tech. Rep., 2021.

J. Ebbers and R. Haeb-Umbach, "Pre-training and self-training for sound event detection in domestic environments," DCASE, Tech. Rep., 2022.

S. Xiao, J. Shen, A. Hu, X. Zhang, P. Zhang, and Y. Yan, "Sound event detection with weak prediction for dcase 2023 challenge task4a," DCASE Tech. Rep., 2023.

W. Zhang, B. Lin, Y. Yan, A. Zhou, Y. Ye, and X. Zhu, "Multi-features fusion for underwater acoustic target recognition based on convolution recurrent neural networks," Proc. IEEE BigDIA, 342-346 (2022).

10.1109/BigDIA56350.2022.9874151

P. Qi, G. Yin, and L. Zhang, "Underwater acoustic target recognition using RCRNN and wavelet-auditory feature," Proc. Multimed. Tools and Appl. 47295-47317 (2024).

10.1007/s11042-023-17406-2

D.-H. Lee, "Pseudo-label: The simple and efficient semi-supervised learning method for deep neural networks," Proc. ICML, 896 (2013).

X. Dong, J. Shen, and L. Shao, "Rethinking clustering-based pseudo-labeling for unsupervised meta-learning," Proc. ECCV, 169-186 (2022).

10.1007/978-3-031-20044-1_10

D. A. Abraham, Underwater Acoustic Signal Processing: Modeling, Detection, and Estimation (Springer, Cham, 2019), pp. 180-189.

10.1007/978-3-319-92983-5

G. Kim and S. Lee, "Reverberation suppression method for active sonar systems using non-negative matrix factorization with pre-trained frequency basis matrix," IEEE Access, 9, 148060-148075 (2021).

10.1109/ACCESS.2021.3124509

D. P. Kingma and J. Ba, "Adam: A method for stochastic optimization," Proc. 3rd ICLR, 1412.6980 (2015).

A. Géron, Hands-on Machine Learning with Scikit-Learn, Keras, and TensorFlow, 2nd ed. (O'Reilly Media, Inc., California, 2022), pp. 93-98.

Information

Publisher :The Acoustical Society of Korea
Publisher(Ko) :한국음향학회
Journal Title :The Journal of the Acoustical Society of Korea
Journal Title(Ko) :한국음향학회지
Volume : 43
No :5
Pages :502-510
Received Date : 2024-05-17
Revised Date : 2024-07-15
Accepted Date : 2024-07-25
DOI :https://doi.org/10.7776/ASK.2024.43.5.502

The Journal of the Acoustical Society of KoreaISSN:1225-4428(Print) 2287-3775(Online)한국음향학회

All Issue