基于一维注意力机制的卷积神经网络水声目标识别

张羽飞; 赵梅; 胡长青; 郭政

doi:10.11993/j.issn.2096-3920.2025-0053

基于一维注意力机制的卷积神经网络水声目标识别

doi: 10.11993/j.issn.2096-3920.2025-0053

张羽飞^{1, 2,},
赵梅^1,,
胡长青^1,,
郭政^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

详细信息

作者简介:
张羽飞 (2000-), 男, 在读硕士, 主要研究方向为水声目标识别

中图分类号: TJ630; U663
计量
- 文章访问数: 34
- HTML全文浏览量: 15
- PDF下载量: 3
- 被引次数: 0
出版历程
- 收稿日期: 2025-04-07
- 修回日期: 2025-05-18
- 录用日期: 2025-05-20
- 网络出版日期: 2025-09-12

Underwater Acoustic Target Recognition Based on One-Dimensional Convolutional Neural Network with Attention Mechanism

ZHANG Yufei^{1, 2
,},
ZHAO Mei^1
,,
HU Changqing^1
,,
GUO Zheng^1
,

1.
Shanghai Acoustics Laboratory, Chinese Academy of Sciences, Shanghai 201815, China
2.
University of Chinese Academy of Sciences, Beijing 100049, China

摘要

摘要: 针对基于深度学习的水声目标识别模型存在网络参数复杂、计算成本高等问题, 提出一种轻量级一维注意力机制卷积神经网络水声目标识别模型。首先, 在特征提取阶段, 选择频谱、梅尔谱、色度、谱对比度和色调特征, 将其重构并融合为一维混合特征; 之后, 通过多尺度残差卷积(MRC)以增强混合特征在不同尺度上的特征表达能力; 同时, 引入卷积注意力模块(CBAM), 通过通道注意力和空间注意力模块自适应地调整特征权重, 提升模型对关键区域的关注。实验结果表明, 该模型在ShipsEar数据集上的平均识别率达到98.58%, 表现出良好的分类效果, 且运算量大大减少。
- 水声目标识别 /
- 多特征融合 /
- 注意力机制 /
- 一维卷积神经网络
Abstract: To address the issues of complex network parameters and high computational costs in deep learning-based underwater acoustic target recognition models, this study proposed a lightweight one-dimensional convolutional neural network with an attention mechanism for underwater acoustic target recognition. First, during the feature extraction stage, spectral, Mel-spectrogram, chroma, spectral contrast, and tonal features were selected and reconstructed into a fused one-dimensional hybrid feature. Next, the hybrid feature was processed by a multi-scale residual convolution(MRC) module to enhance feature representation across different scales. Simultaneously, a convolutional block attention module(CBAM) was introduced to adaptively adjust feature weights through channel and spatial attention modules, improving the model’s focus on critical regions. Experimental results show that the proposed model achieves an average recognition accuracy of 98.58% on the ShipsEar dataset, demonstrating excellent classification performance and significantly reducing computational complexity.
- underwater acoustic target recognition /
- multi-feature fusion /
- attention mechanism /
- one-dimensional convolutional neural network

HTML全文

图 1 CBAM模型

Figure 1. Model of the CBAM

下载: 全尺寸图片幻灯片

图 2 1D-MRC-CBAM模型

Figure 2. 1D-MRC-CBAM model

下载: 全尺寸图片幻灯片

图 3 卷积块的具体结构

Figure 3. The specific structure of a convolutional block

下载: 全尺寸图片幻灯片

图 4 识别算法流程图

Figure 4. Flowchart of the recognition algorithm

下载: 全尺寸图片幻灯片

图 5 ShipsEar数据集训练损失与识别率曲线

Figure 5. Training loss and recognition rates curves on the ShipsEar dataset

下载: 全尺寸图片幻灯片

图 6 测试集输出可视化结果

Figure 6. Visualization results of the test set outputs

下载: 全尺寸图片幻灯片

图 7 移除CBAM模块后模型训练曲线

Figure 7. Training curves of the model without CBAM module

下载: 全尺寸图片幻灯片

图 8 不同卷积核尺寸下的验证集损失与识别率

Figure 8. Validation loss and recognition rates under different convolutional kernel sizes

下载: 全尺寸图片幻灯片

图 9 1D-MRC-CBAM模型在东海数据上的混淆矩阵

Figure 9. Confusion matrix of the 1D-MRC-CBAM model on the East China Sea measured data

下载: 全尺寸图片幻灯片

表 1 各模块输出大小及运算量

Table 1. Output size and computational volumes of each module

模块	输入大小	输出大小	运算量
A_Block	(1, 1, 2 000)	(1, 16, 500)	2.5×10³
B_Block	(1, 1, 2 000)	(1, 16, 500)	4.1×10³
C_Block	(1, 1, 2 000)	(1, 16, 500)	5.7×10³
拼接层	(1, 16, 31)	(1, 48, 31)	12.3×10³
D_Block	(1, 48, 31)	(1, 16, 31)	4.3×10³
FC1	(1, 16, 31)	(1, 1, 256)	127×10³
FC2	(1, 1, 256)	(1, 1, 16)	54.1×10³

下载: 导出CSV

表 2 ShipsEar数据集样本分类

Table 2. Sample classification of the ShipsEar dataset

分类	船只种类	样本数
A	挖泥船/渔船/贻贝船/拖网渔船/拖船	1 875
B	摩托艇/引航船/帆船	1 560
C	客船	4 270
D	邮轮/滚装船	2 455
E	自然噪声	1 140

下载: 导出CSV

表 3 不同特征重构前后维度大小

Table 3. Dimensional sizes of features before and after reconstruction

特征	原始大小	重构后大小
FFT	1×1 000	1×1 000
MFCC	25×20	1×500
Chorma	12×20	1×240
Contrast	6×20	1×120
Tonnetz	6×20	1×120

下载: 导出CSV

表 4 模型参数设置

Table 4. Parameterization of the model

参数选择	参数设置
学习率	10⁻³
迭代次数	100
优化器	Adam
损失函数	交叉熵损失函数
训练批次	64

下载: 导出CSV

表 5 不同信噪比下模型识别率

Table 5. Recognition rates of the model under different signal-to-noise ratios

信噪比/dB	识别率/%
信噪比/dB	A	B	C	D	E
5	97.01	95.88	97.28	99.18	100
0	93.40	93.03	95.62	98.74	97.87
−5	96.39	90.41	93.03	94.67	94.06
−10	88.06	77.70	85.45	95.45	93.85
−15	77.90	70.80	78.99	82.37	87.00
−20	64.24	52.94	74.07	66.80	80.86

下载: 导出CSV

表 6 不同输入特征下模型识别率

Table 6. Model recognition rates with different input features

输入特征	识别率/%
FFT	97.32
MFCC	97.43
FFT+MFCC	98.32
FMCCT	98.58

下载: 导出CSV

表 7 不同模型识别率和参数量对比

Table 7. Comparison of recognition rates and parameters of different methods

模型	输入特征	识别率/%	参数量
文献[20]	Mel频谱	96.4	0.47×10⁶
文献[21]	SSA	98.6	0.26×10⁶
文献[22]	STFT谱	99.4	5.47×10⁶
文献[23]	Log-Mel序列	98.5	3.5×10⁶
文献[24]	原始信号	99.8	0.61×10⁶
1D-MRC-CBAM	FMCCT混合特征	98.58	0.2×10⁶

下载: 导出CSV

参考文献(24)

[1]	徐及, 黄兆琼, 李琛, 等. 深度学习在水下目标被动识别中的应用进展[J]. 信号处理, 2019, 35(9): 1460-1475. XU J, HUANG Z Q, LI C, et al. Advances in applications of deep learning in passive underwater target recognition[J]. Signal Processing, 2019, 35(9): 1460-1475.
[2]	蔡悦斌, 张明之, 史习智, 等. 舰船噪声波形结构特征提取及分类研究[J]. 电子学报, 1999(6): 129-130. doi: 10.3321/j.issn:0372-2112.1999.06.030 CAI Y B, ZHANG M Z, SHI X Z, et al. Research on extraction and classification of structural features of ship-radiated noise waveform[J]. Acta Electronica Sinica, 1999(6): 129-130. doi: 10.3321/j.issn:0372-2112.1999.06.030
[3]	葛轶洲, 姚泽, 张歆, 等. 水声目标的MFCC特征提取与分类识别[J]. 计算机仿真, 2024, 41(2): 13-16, 33. doi: 10.3969/j.issn.1006-9348.2024.02.003 GE Y Z, YAO Z, ZHANG X, et al. MFCC feature extraction and classification recognition of underwater acoustic targets[J]. Computer Simulation, 2024, 41(2): 13-16, 33. doi: 10.3969/j.issn.1006-9348.2024.02.003
[4]	谢东日. 水声目标非线性特征提取研究[D]. 厦门: 厦门大学, 2020.
[5]	倪俊帅, 赵梅, 胡长青. 基于深度学习的舰船辐射噪声多特征融合分类[J]. 声学技术, 2020, 39(3): 366-371. NI J S, ZHAO M, HU C Q. Multi-feature fusion classification of ship-radiated noise based on deep learning[J]. Technical Acoustics, 2020, 39(3): 366-371.
[6]	梁喆, 侯朋, 夏春艳, 等. 融合时频域特征的舰船识别方法及实验研究[J]. 声学技术, 2021, 40(5): 607-613. LIANG Z, HOU P, XIA C Y, et al. Ship recognition method and experimental study based on fusion of time-frequency domain features[J]. Technical Acoustics, 2021, 40(5): 607-613.
[7]	刘承伟, 洪峰, 冯海泓, 等. 结合多尺度卷积网络和双端注意力机制的水声目标识别[J]. 声学技术, 2023, 42(2): 161-167. LIU C W, HONG F, FENG H H, et al. Underwater acoustic target recognition combining multi-scale convolutional networks and dual attention mechanism[J]. Technical Acoustics, 2023, 42(2): 161-167.
[8]	杨基睿, 鄢社锋, 曾迪, 等. 改进通道注意力机制的时域水声信号识别网络[J]. 信号处理, 2023, 39(6): 1025-1035. YANG J R, YAN S F, ZENG D, et al. Time-domain underwater acoustic signal recognition network with improved channel attention mechanism[J]. Signal Processing, 2023, 39(6): 1025-1035.
[9]	LIU Y X, ZHANG B Q, KONG F T, et al. Underwater acoustic classification using wavelet scattering transform and convolutional neural network with limited dataset[J]. Applied Acoustics, 2025, 232: 110564. doi: 10.1016/j.apacoust.2025.110564
[10]	LYU C G, HU X Y, NIU Z H, et al. A light-weight neural network for marine acoustic signal recognition suitable for fiber-optic hydrophones[J]. Expert Systems with Applications, 2024, 235: 121235.
[11]	杨路飞, 章新华, 吴秉坤, 等. 基于MFCC特征的被动水声目标深度学习分类方法[J]. 舰船科学技术, 2020, 42(19): 129-133. doi: 10.3404/j.issn.1672-7649.2020.10.025 YANG L F, ZHANG X H, WU B K, et al. Passive underwater acoustic target classification method based on MFCC features and deep learning[J]. Ship Science and Technology, 2020, 42(19): 129-133. doi: 10.3404/j.issn.1672-7649.2020.10.025
[12]	ZHANG S K, TIAN D Y, WANG C, et al. Intelligent recognition of underwater acoustic target noise on underwater glider platform[C]//2018 Chinese Automation Congress(CAC). Piscataway, NJ: IEEE, 2018: 4189-4193.
[13]	任晨曦, 王黎明, 韩星程, 等. 基于联合神经网络的水声目标识别方法[J]. 舰船科学技术, 2022, 44(1): 136-141. doi: 10.3404/j.issn.1672-7649.2022.01.026 REN C X, WANG L M, HAN X C, et al. Underwater acoustic target recognition method based on joint neural network[J]. Ship Science and Technology, 2022, 44(1): 136-141. doi: 10.3404/j.issn.1672-7649.2022.01.026
[14]	YANG K, WANG B, FANG Z, et al. An end-to-end underwater acoustic target recognition model based on one-dimensional convolution and transformer[J]. Journal of Marine Science and Engineering. 2024; 12(10): 1793.
[15]	LIU D Y, YANG H Y, HOU W M, et al. A novel underwater acoustic target recognition method based on MFCC and RACNN[J]. Sensors, 2024, 24(1): 273. doi: 10.3390/s24010273
[16]	YÜCESOY E. Gender recognition based on the stacking of different acoustic features[J]. Applied Sciences, 2024, 14(15): 6564. doi: 10.3390/app14156564
[17]	ZHANG S, GAO Y, CAI J, et al. A novel bird sound recognition method based on multi-feature fusion and a transformer encoder[J]. Sensors, 2023, 23(19): 8099. doi: 10.3390/s23198099
[18]	WOO S, PARK J, LEE J Y, et al. CBAM: Convolutional block attention module[J]. European Conference on Computer Vision, 2018, 1: 3-19.
[19]	DAVID S D, TORRES G S, ANTONIO C L, et al. ShipsEar: An underwater vessel noise database[J]. Applied Acoustics, 2016, 113: 64-69.
[20]	YANG S, JIN A, ZENG X, et al. Underwater acoustic target recognition based on sub-band concatenated Mel spectrogram and multidomain attention mechanism[J]. Engineering Applications of Artificial Intelligence, 2024, 133: 107983. doi: 10.1016/j.engappai.2024.107983
[21]	NI J, JI F, LU S, et al. Underwater target recognition method based on singular spectrum analysis and channel attention convolutional neural network[J]. Sensors, 2025, 25: 2573. doi: 10.3390/s25082573
[22]	WANG B, ZHANG W, ZHU Y, et al. An underwater acoustic target recognition method based on AMNet[J]. IEEE Geoscience and Remote Sensing Letters, 2023, 20: 1-5.
[23]	WANG J Y, QIAN P, CHEN Y X, et al. Adaptive underwater acoustic target recognition based on multi-scale residual and attention mechanism[J]. Digital Signal Processing, 2025, 163: 105193. doi: 10.1016/j.dsp.2025.105193
[24]	NI J, JI F, LU S, et al. An auditory convolutional neural network for underwater acoustic target timbre feature extraction and recognition[J]. Remote Sens, 2024, 16: 3074.