联合自适应滤波-深度学习的舰船辐射噪声线谱增强

蔡婷婷; 俞孙泽; 赵梅

doi:10.11993/j.issn.2096-3920.2025-0040

联合自适应滤波-深度学习的舰船辐射噪声线谱增强

doi: 10.11993/j.issn.2096-3920.2025-0040

蔡婷婷^{1, 2,},
俞孙泽^3,,
赵梅^1,

1.
中国科学院声学研究所东海研究站, 上海, 201815
2.
中国科学院大学, 北京, 100049
3.
海军装备部驻上海地区第七军事代表室, 上海, 201108

详细信息

作者简介:
蔡婷婷(2001-), 女, 在读硕士, 主要研究方向为水下目标线谱增强与检测

中图分类号: TJ6; U674.941
计量
- 文章访问数: 48
- HTML全文浏览量: 23
- PDF下载量: 12
- 被引次数: 0
出版历程
- 收稿日期: 2025-03-03
- 修回日期: 2025-03-17
- 录用日期: 2025-03-24
- 网络出版日期: 2025-07-14

Ship Radiated Noise Line Spectrum Enhancement Based on Adaptive Filtering-Deep Learning Fusion

CAI Tingting^{1, 2
,},
YU Sunze^3
,,
ZHAO Mei^1
,

1.
Shanghai Acoustics Laboratory, Chinese Academy of Sciences, Shanghai 201815, China
2.
University of Chinese Academy of Sciences, Beijing 100049, China
3.
The Seventh Military Representative Office of Naval Equipment Department in Shanghai, Shanghai 201108, China

摘要

摘要: 针对复杂水声环境中舰船辐射噪声线谱特征难以辨识的问题, 提出了联合自适应滤波与深度学习(DL)的线谱增强模型。该模型融合双层协同自适应滤波模块与DL模块, 构建多层级特征增强框架。自适应滤波模块结合频域自适应滤波(FDAF)与最大相关熵准则(MCC), 在抑制宽带背景噪声的同时提高了对非平稳噪声的抑制能力。DL模块采用双向长短时记忆网络(BiLSTM)提取线谱的局部时间依赖关系, 结合注意力残差机制增强对弱线谱的关注, 并利用Transformer 编码器捕捉时频域长程关联。该模型充分结合滤波和DL的优点, 既有效抑制噪声, 又提升了对弱线谱的增强效果。仿真验证结果表明, 该方法在整体线谱增强及弱线谱强化方面均优于单一自适应滤波或DL模型。湖试数据验证进一步证实了该方法的有效性。
- 舰船辐射噪声 /
- 线谱增强 /
- 深度学习 /
- 自适应滤波 /
- Transformer 编码器 /
- 注意力机制
Abstract: To address the challenge of identifying line spectrum features of ship radiated noise in complex underwater acoustic environments, a line spectrum enhancement model combining adaptive filtering and deep learning(DL) was proposed. This model integrated a double-layer collaborative adaptive filtering module with a DL module to form a multi-level feature enhancement framework. The adaptive filtering module combined frequency-domain adaptive filtering(FDAF) with the maximum correntropy criterion(MCC) to enhance the suppression of non-stationary noise while effectively reducing broadband background noise. The DL module employed a bi-directional long short-term memory(BiLSTM) network to extract the local temporal dependencies of line spectra. It also incorporated an attention residual mechanism to focus on weak line spectra and utilized a Transformer encoder to capture long-range correlations in the time-frequency domain. The model effectively combined the advantages of filtering and DL, both suppressing noise and enhancing the detection of weak line spectra. Simulation results demonstrate that the proposed method outperforms single adaptive filtering or DL models in terms of overall line spectrum enhancement and weak line spectrum enhancement. The effectiveness of this method is further validated through lake test data.
- ship radiated noise /
- line spectrum enhancement /
- deep learning /
- adaptive filtering /
- Transformer encoder /
- attention mechanism

HTML全文

图 1 MCC-FDAF-DL模型技术路线图

Figure 1. Roadmap of technical approach for MCC-FDAF-DL model

下载: 全尺寸图片幻灯片

图 2 FDAF基本框架图

Figure 2. Basic framework of FDAF

下载: 全尺寸图片幻灯片

图 3 MCC-FDAF-DL线谱增强模型结构

Figure 3. Line spectrum enhanced model structure of MCC-FDAF-DL

下载: 全尺寸图片幻灯片

图 4 BiLSTM模型结构

Figure 4. Structure of BiLSTM model

下载: 全尺寸图片幻灯片

图 5 SAR模块结构

Figure 5. Structure of SAR module

下载: 全尺寸图片幻灯片

图 6 Transformer编码器结构

Figure 6. Structure of Transformer encoder

下载: 全尺寸图片幻灯片

图 7 多头自注意力机制结构

Figure 7. Structure of multi-head self-attention

下载: 全尺寸图片幻灯片

图 8 仿真信号波形图

Figure 8. Waveforms of simulation signals

下载: 全尺寸图片幻灯片

图 9 SNR为−15 dB时单一-联合方法输出信号LOFAR谱

Figure 9. LOFAR spectrum of the output signals of the single and joint method at SNR is −15 dB

下载: 全尺寸图片幻灯片

图 10 SNR为−5 dB时单一-联合方法输出信号LOFAR谱

Figure 10. LOFAR spectrum of the output signals of the single and joint method at SNR is −5 dB

下载: 全尺寸图片幻灯片

图 11 湖试方案

Figure 11. Program of the lake test

下载: 全尺寸图片幻灯片

图 12 湖试数据输出信号LOFAR谱

Figure 12. LOFAR spectrum of the output signals of the lake test data

下载: 全尺寸图片幻灯片

图 13 0~200 Hz区间输出信号LOFAR谱

Figure 13. LOFAR spectrum of the output signals in the 0-200 Hz range

下载: 全尺寸图片幻灯片

图 14 输入信号75 Hz处线谱信号缺失示意图

Figure 14. Line spectrum signal missing of input signal at 75 Hz

下载: 全尺寸图片幻灯片

表 1 模型参数设置

Table 1. Parameter setting of the model

模块	参数名称	参数值
自适应滤波模块	滤波器阶数	2 000
自适应滤波模块	步长	0.000 01
BiLSTM	隐藏层数	300
Transformer编码器	编码器层数	2
全局参数	损失函数	SmoothL1Loss
	优化器	Adam
	迭代次数	300
	学习率	0.001

下载: 导出CSV

表 2 仿真线谱信号参数设置

Table 2. Parameters setting of the simulated line spectrum signals

序号	幅值	频率/Hz
1	3.5	100
2	2	200
3	5	500
4	6	1 500

下载: 导出CSV

表 3 消融实验结果

Table 3. Results of ablation experiments

模型	SNR/dB
模型	−5	−7	−10	−12	−15
MCC-FDAF-DL	0.957	0.951	0.944	0.938	0.928
BiLSTM	0.770	0.763	0.757	0.746	0.741
BiLSTM + SAR	0.845	0.839	0.832	0.825	0.815
BiLSTM + Transform编码器	0.846	0.839	0.834	0.821	0.808

下载: 导出CSV

参考文献(18)

[1]	AINSLIE M A. Principles of sonar performance modeling[M]. Berlin: Springer, 2010.
[2]	唐冰钊. 基于熵和模态分解的舰船辐射噪声特征提取方法研究[D]. 西安: 西安理工大学, 2025.
[3]	WIDROW B, GLOVER J R, MCCOOL J M, et al. Adaptive noise cancelling: Principles and applications[J]. Proceedings of the IEEE, 1975, 63(12): 1692-1716. doi: 10.1109/PROC.1975.10036
[4]	刘辉涛, 丛卫华, 潘翔. 窄带弱信号的线谱检测——相干累加频域批处理自适应线谱增强方法[J]. 浙江大学学报(工学版), 2007(12): 2048-2051. LIU H T, CONG W H, PAN X. Line spectral detection of tone weak signal-an adaptive line enhancement technique using coherent addition and frequency domain batch[J]. Journal of Zhejiang University(Engineering Science), 2007(12): 2048-2051.
[5]	HAO Y, CHI C, LIANG G. Sparsity-driven adaptive enhancement of underwater acoustic tonals for passive sonars[J]. The Journal of the Acoustical Society of America, 2020, 147(4): 2192-2204. doi: 10.1121/10.0001017
[6]	罗斌, 王茂法, 王世闯. 一种高效的弱目标线谱检测算法[J]. 声学技术, 2017, 36(2): 171-176. LUO B, WANG M F, WANG S C. A highly efficient weak target line-spectrum detection algorithm[J]. Technical Acoustics, 2017, 36(2): 171-176.
[7]	HAO Q, ZHANG X, WANG Y, et al. A novel rail defect detection method based on undecimated lifting wavelet packet transform and shannon entropy-improved adaptive line enhancer[J]. Journal of Sound and Vibration, 2018, 425: 208-220. doi: 10.1016/j.jsv.2018.04.003
[8]	王燕, 上官佩熙, 郝宇, 等. 非高斯噪声背景下的目标辐射线谱自适应增强方法[J]. 声学学报, 2024, 49(5): 927-938. doi: 10.12395/0371-0025.2023040 WANG Y, SHANGGUAN P X, HAO Y, et al. Adaptive enhancer of the target radiated line-spectrum under non-Gaussian noise[J]. Acta Acustica, 2024, 49(5): 927-938. doi: 10.12395/0371-0025.2023040
[9]	张奇, 笪良龙, 王超, 等. 基于深度学习的水声被动目标识别研究综述[J]. 电子与信息学报, 2023, 45(11): 4190-4202. ZHANG Q, DA L L, WANG C, et al. An overview on underwater acoustic passive target recognition based on deep learning[J]. Journal of Electronics & Information Technology, 2023, 45(11): 4190-4202.
[10]	李悦, 马晓川, 王磊, 等. 非高斯环境下的深度学习脉冲信号去噪与重构[J]. 应用声学, 2021(1): 131-141. doi: 10.11684/j.issn.1000-310X.2021.01.015 LI Y, MA X C, WANG L, et al. Using deep learning to de-noise and reconstruct pulse signals in non-Gaussian environment[J]. Journal of Applied Acoustics, 2021(1): 131-141. doi: 10.11684/j.issn.1000-310X.2021.01.015
[11]	JU D, CHI C, LI Z, et al. Deep-learning-based line enhancer for passive sonar systems[J]. IET Radar, Sonar & Navigation, 2022, 16(3): 589-601.
[12]	杨路飞, 章新华, 吴秉坤. 基于长短时记忆网络的被动声纳目标信号LOFAR谱增强研究[J]. 电声技术, 2020, 44(6): 101-103. YANG L F, ZHANG X H, WU B K. A study on signal LOFAR spectrum enhancement of passive sonar target based on short and short time memory network[J]. Audio Engineering, 2020, 44(6): 101-103.
[13]	古天龙, 张清智, 李晶晶. 基于时-频注意力机制网络的水声目标线谱增强[J]. 电子与信息学报, 2024, 46(1): 92-100. GU T L, ZHANG Q Z, LI J J. Line spectrum enhancement of underwater acoustic targets based on a time-frequency attention network[J]. Journal of Electronics & Information Technology, 2024, 46(1): 92-100.
[14]	HE T, FENG S, YANG J, et al. Underwater acoustic signal LOFAR spectrogram denoising based on enhanced simulation[J]. Applied Sciences, 2024, 14(23): 10931. doi: 10.3390/app142310931
[15]	LEI L, SHAO S, LIANG L. An evolutionary deep learning model based on EWKM, random forest algorithm, SSA and BiLSTM for building energy consumption prediction[J]. Energy, 2024, 288: 129795. doi: 10.1016/j.energy.2023.129795
[16]	SHERSTINSKY A. Fundamentals of recurrent neural network(RNN) and long short-term memory(LSTM) network[J]. Physica D: Nonlinear Phenomena, 2020, 404: 132306. doi: 10.1016/j.physd.2019.132306
[17]	WANG Y, QIU S, HU G, et al. Suppressing short time marine ambient noise based on deep complex unet to enhance the vessel radiation signal in LOFAR spectrogram[J]. Journal of Applied Geophysics, 2025, 233: 105611. doi: 10.1016/j.jappgeo.2024.105611
[18]	焦晨光, 张小波. 一种改进的基于结构相似性的非局部均值图像去噪算法[J]. 智能计算机与应用, 2025(2): 17-23. JIAO C G, ZHANG X B. Based on structural similarity improved non-local means image denoising algorithm[J]. Intelligent Computer and Applications, 2025(2): 17-23.