Application of a multilayer perceptron with adaptive learning for the analysis of acoustic signals in the water environment

DOI: 10.31673/2412-9070.2023.039000

Abstract

Modern studies of acoustic, magnetic, and other fields of the marine environment are conducted under conditions that are characterized by a high rate of change in the external environment and an ever-increasing amount of information received on various channels in real time. This is especially typical for studies of signals with passive reception methods, which are characterized by a low SNR. Additionally, in shallow sea areas with a significant number of signal sources and serious interference, passive methods of signal monitoring encounter difficulties in noise measurement and the problem of low accuracy. Experience from such studies involving automatic information processing reveals that classical theories are often not applicable. The most promising field of science that can efficiently address current challenges is machine learning. Machine learning offers various tools, such as neural networks, for the task of signal analysis. There are numerous variations in this field, and there are many possible solutions. However, not all solutions work well for specific tasks with various data. Hence, there is a need to enhance existing methods that can address recent challenges. A new method has been proposed, which is more efficient and provides better results in the field of underwater signal analysis. In this study, a multilayer perceptron with batch normalization and adaptive learning was proposed. These solutions surpassed the original method and demonstrated their ability to efficiently analyze data from underwater acoustic vessel sources. The accuracy of the proposed classifier reached up to 96% on a dataset containing 4 vessels with various SNR levels.

Keywords: machine learning; acoustic signals; underwater environment; classification; multilayer perceptron; adaptive learning.

References
1. Doan V.-S., Huynh-The T., Kim D.-S. Underwater acoustic target classification based on dense convolutional neural network // IEEE Geosci. Remote Sens. Lett., Oct. 2020.
2. Underwater target recognition based on multidecision lofar spectrum enhancement: a deep learning approach / J. Chen, J. Liu, C. Liu [et al.] // arXiv preprint arXiv:2104.12362 (2021)
3. Han X. C., Ren C., Wang L., Bai Y. Underwater acoustic target recognition method based on a joint neural network // PLoS ONE 2022. 17(4): e0266425.
4. Convolutional Neural Network With Second Order Pooling for Underwater Target Classification / X. Cao, R. Togneri, X. Zhang, Y. Yu // IEEE Sensors Journal, 15 April15, 2019. vol. 19, no. 8. P. 3058–3066,
5. A new cooperative deep learning method for underwater acoustic target recognition / Yang Honghui, Xu Guanghui, Yi Shuzhen, Li Yiqing // OCEANS 2019-Marseille / IEEE. 2019. P. 1–4.
6. Underwater target classification at greater depths using deep neural network with joint multipledomain feature / Cao Xu, Zhang Xiaomin, Togneri Roberto, Yu Yang // IET Radar, Sonar & Navigation. 2019. Vol. 13, no. 3. P. 484–491.
7. DeepShip: An underwater acoustic benchmark dataset and a separable convolution based autoencoder for classification / Muhammad Irfan, Zheng Jiangbin, Shahid Ali, Muhammad Iqbal, Zafar Masood, Umar Hamid // Expert Systems with Applications, 2021. Vol. 183, 115270, ISSN 0957-4174
8. Santos-Domínguez David, Torres-Guijarro Soledad, Cardenal-Lopez Antonio, Pena-Gimenez Antonio. ShipsEar: An underwater vessel noise database. Applied Acoustics 2016;113:64–9