基于混合扩散模型的水声数据增扩方法

doi:10.12305/j.issn.1001-506X.2026.09.02

系统工程与电子技术 ›› 2026, Vol. 48 ›› Issue (9): 2898-2906.doi: 10.12305/j.issn.1001-506X.2026.09.02

• 电子技术 • 上一篇

基于混合扩散模型的水声数据增扩方法

马治勋¹^,²(), 汤宁¹(), 李璇¹^,²(), 郝程鹏¹^,²()

1. 中国科学院声学研究所，北京 100190
2. 中国科学院大学电子电气与通信工程学院，北京 100049

收稿日期:2025-09-11 修回日期:2025-11-10 接受日期:2025-12-02 出版日期:2026-01-20 发布日期:2026-01-20
通讯作者: 郝程鹏 E-mail:mazhixun@mail.ioa.ac.cn;tangning@mail.ioa.ac.cn;lixuan@mail.ioa.ac.cn;haochengp@mail.ioa.ac.cn
作者简介:马治勋（1991—），男，助理研究员，硕士，主要研究方向为水声信号处理、自适应目标检测与目标识别
汤　宁（1994—），女，助理研究员，博士，主要研究方向为水声信号处理
李　璇（1983—），女，研究员，博士，主要研究方向为水声信号处理、目标识别

Mixed diffusion model-based underwater acoustic data augmentation method

Zhixun Ma¹^,²(), Ning Tang¹(), Xuan Li¹^,²(), Chengpeng Hao¹^,²()

1. Institute of Acoustics，Chinese Academy of Sciences，Beijing 100190，China
2. School of Electronic，Electrical and Communication Engineering，University of Chinese Academy of Sciences，Beijing 100049，China

Received:2025-09-11 Revised:2025-11-10 Accepted:2025-12-02 Online:2026-01-20 Published:2026-01-20
Contact: Chengpeng Hao E-mail:mazhixun@mail.ioa.ac.cn;tangning@mail.ioa.ac.cn;lixuan@mail.ioa.ac.cn;haochengp@mail.ioa.ac.cn

摘要/Abstract

摘要：

针对水声目标识别领域广泛存在的数据稀缺性问题，提出一种基于混合扩散模型的数据增扩方法。为区别于传统频谱变换或生成对抗网络方法，将扩散模型引入水声数据增扩领域，通过构建合成样本集有效提升了现有深度学习模型的泛化能力；提出一种基于扩散模型的混合生成策略，通过预先混合类别标签与数据，扩展数据生成空间，从而进一步降低过拟合风险。在两个公开水声数据集上的实验表明，基于混合扩散模型的数据增扩方法能够合成时频结构逼真的水声数据，其中视觉 Transformer 模型在ShipsEar数据集上准确率最高提升14.6%，有效提高了多种识别模型的分类性能。

关键词: 数据增扩, 水声目标识别, 扩散模型, 深度学习

Abstract:

In addressing the pervasive data scarcity problem in the field of underwater acoustic target recognition, a data augmentation method is proposed based on hybrid diffusion model. Distinguishing itself from conventional frequency spectral transformations or generative adversarial networks methods, the diffusion model is introduced into the field of underwater acoustic data augmentation. By constructing a synthetic sample set, the generalization ability of existing deep learning models is effectively improved. A hybrid generation strategy based on diffusion models is proposed, which expands the data generation space by pre-mixing category labels with data, thereby further reducing the risk of overfitting. Experiments on two public underwater acoustic datasets demonstrate that the data augmentation method based on the hybrid diffusion model can synthesize underwater acoustic data with realistic time-frequency structures. Specifically, the vision Transformer model achieves an accuracy improvement of up to 14.6% on the ShipsEar dataset, effectively enhancing the classification performance of various recognition models.

Key words: data augmentation, underwater acoustic target recognition, diffusion model, deep learning

中图分类号:

马治勋, 汤宁, 李璇, 郝程鹏. 基于混合扩散模型的水声数据增扩方法[J]. 系统工程与电子技术, 2026, 48(9): 2898-2906.

Zhixun Ma, Ning Tang, Xuan Li, Chengpeng Hao. Mixed diffusion model-based underwater acoustic data augmentation method[J]. Systems Engineering and Electronics, 2026, 48(9): 2898-2906.

图/表 10

图1

图2

表1

表2

表3

图3

图4

图5

图6

图7

参考文献 30

1	Fillinger L, De-Theije P, Zampolli M, et al. Towards a passive acoustic underwater system for protecting harbors against intruders[C]//International Waterside Security Conference, 2010.
2	Chung K W, Sutin A, Sedunov A, et al. DEMON acoustic ship signature measurements in an urban harbor[J]. Advances in Acoustics and Vibration, 2011, 1, 952798. doi: 10.1155/2011/952798
3	Mishachandar B, Vairamuthu S. Diverse ocean noise classification using deep learning[J]. Applied Acoustics, 2021, 181, 108141. doi: 10.1016/j.apacoust.2021.108141
4	Kheddar H, Hemis M, Himeur Y. Automatic speech recognition using advanced deep learning approaches: a survey[J]. Information Fusion, 2024, 109, 102422. doi: 10.1016/j.inffus.2024.102422
5	张赵管. 机器学习算法在音频信号降噪中的应用研究[J]. 电声技术, 2024, 48 (11): 39. doi: 10.16311/j.audioe.2024.11.013
6	Ren J W, Xie Y, Zhang X W, et al. UALF: a learnable front-end for intelligent underwater acoustic classification system[J]. Ocean Engineering, 2022, 264, 112394. doi: 10.1016/j.oceaneng.2022.112394
7	Xie Y, Ren J W, Xu J. Adaptive ship-radiated noise recognition with learnable fine-grained wavelet transform[J]. Ocean Engineering, 2022, 265, 112626. doi: 10.1016/j.oceaneng.2022.112626
8	Wu F, Yao H Y, Zhao Z D, et al. ESTMST-ST: an end-to-end soft threshold and multi-loss self-distillation based Swin-Transformer for underwater acoustic signal recognition[J]. IEEE Trans. on Geoscience and Remote Sensing, 2024, 63, 1. doi: 10.1109/tgrs.2024.3520860
9	于学洋, 迟骋, 李淑秋, 等. 联合线谱增强与深度神经网络的水声目标识别[J]. 声学学报, 2024, 49 (4): 656. doi: 10.12395/0371-0025.2024021
10	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. doi: 10.11999/JEIT221301
11	Liu F, Shen T S, Luo Z L, et al. Underwater target recognition using convolutional recurrent neural networks with 3-D mel-spectrogram and data augmentation[J]. Applied Acoustics, 2021, 178, 107989. doi: 10.1016/j.apacoust.2021.107989
12	Park D S, Zhang Y, Chiu C C, et al. SpecAugment on large scale datasets[C]//IEEE International Conference on Acoustics, Speech and Signal Processing, 2020: 6879.
13	Li D H, Liu F, Shen T S, et al. Data augmentation method for underwater acoustic target recognition based on underwater acoustic channel modeling and transfer learning[J]. Applied Acoustics, 2023, 208, 109344. doi: 10.1016/j.apacoust.2023.109344
14	Zhang H Y, Cisse M, Dauphin Y N, et al. Mixup: beyond empirical risk minimization[C]//International Conference on Learning Representations, 2017.
15	Yun S, Han D, Oh S J, et al. CutMix: regularization strategy to train strong classifiers with localizable features[C]//IEEE/CVF International Conference on Computer Vision, 2019: 6023.
16	Xu J, Xie Y, Wang W C. Underwater acoustic target recognition based on smoothness-inducing regularization and spectrogram-based data augmentation[J]. Ocean Engineering, 2023, 281, 114926. doi: 10.1016/j.oceaneng.2023.114926
17	Dabouei A, Soleymani S, Taherkhani F, et al. SuperMix: supervising the mixing data augmentation[C]//IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2021: 13794.
18	Yang H B, Gu H, Yin J Y, et al. GAN-based sample expansion for underwater acoustic signal[C]//International Conference on Intelligent Computing and Signal Processing, 2020.
19	Chandran C S, Kamal S, Mujeeb A, et al. Generative adversarial learning for improved data efficiency in underwater target classification[J]. Engineering Science and Technology, an International Journal, 2022, 30, 101043. doi: 10.1016/j.jestch.2021.07.006
20	Gong Y P, Huang L Q, Chen L F. Eliminate deviation with deviation for data augmentation and a general multi-modal data learning method[EB/OL]. [2025-08-08]. https://arxiv.org/abs/2101.08533.
21	Fang H Y, Han B R, Zhang S, et al. Data augmentation for object detection via controllable diffusion models[C]//IEEE/CVF Winter Conference on Applications of Computer Vision, 2024: 1257.
22	Santos-Dominguez D, Torres-Guijarro S, Cardenal López A, et al. ShipsEar: an underwater vessel noise database[J]. Applied Acoustics, 2016, 113, 64. doi: 10.1016/j.apacoust.2016.06.008
23	Irfan M, Zheng J B, Ali S, et al. DeepShip: an underwater acoustic benchmark dataset and a separable convolution based autoencoder for classification[J]. Expert Systems with Applications, 2021, 183, 115270. doi: 10.1016/j.eswa.2021.115270
24	Zhang L M, Rao A Y, Agrawala M. Adding conditional control to text-to-image diffusion models[C]//IEEE/CVF International Conference on Computer Vision, 2023: 3836.
25	Wallace B, Gokul A, Ermon S, et al. End-to-end diffusion latent optimization improves classifier guidance[C]//IEEE/CVF International Conference on Computer Vision, 2023: 7280.
26	Ho J, Salimans T. Classifier-free diffusion guidance[C]//Neural Information Processing Systems Workshop, 2021.
27	He K M, Zhang X Y, Ren S Q, et al. Deep residual learning for image recognition[C]//IEEE Conference on Computer Vision and Pattern Recognition, 2016: 770.
28	Simonyan K, Zisserman A. Very deep convolutional networks for large-scale image recognition[EB/OL]. [2025-08-08]. https://arxiv.org/abs/1409.1556.
29	Dosovitskiy A, Beyer L, Kolesnikov A, et al. An image is worth 16×16 words: transformers for image recognition at scale[C]//International Conference on Learning Representations, 2021.
30	Liu Z, Mao H Z, Wu C Y, et al. A ConvNet for the 2020s[C]//IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2022: 11976.

数据集	训练集	测试集	合成数据	总数
ShipsEar	1 848	854	1 848	4 550
DeepShip	32 393	7 429	32 393	72 215

方法	ShipsEar数据集准确率						DeepShip数据集准确率
方法	A类	B类	C类	D类	E类	OA	货船	客船	油轮	拖船	OA
ResNet18	47.4	58.4	85.2	87.5	99.8	77.9	64.8	74.5	74.9	79.1	73.1
+ Cond-Diff	52.6	69.7	89.8	84.3	93.6	80.0	68.3	75.3	72.4	82.8	74.3
+ Mix-Gen	44.9	76.4	89.8	92.6	98.2	81.9	66.7	76.4	76.3	85.0	75.7
VGG16	54.5	74.2	84.9	84.3	92.7	79.0	63.2	74.5	74.1	75.6	71.7
+ Cond-Diff	60.3	51.7	82.0	90.7	92.7	78.4	65.2	74.8	71.6	78.4	72.2
+ Mix-Gen	60.3	58.4	93.7	88.9	96.3	83.0	66.0	70.7	75.8	79.7	72.8
ViT	28.2	24.7	88.7	70.4	91.7	66.7	64.7	67.8	67.8	76.8	68.9
+ Cond-Diff	44.2	67.4	86.6	74.1	91.6	74.3	65.9	64.9	72.0	77.8	69.5
+ Mix-Gen	53.8	75.3	90.8	88.4	86.2	81.3	67.0	73.1	71.3	79.1	72.3
CNN	42.9	68.5	84.2	83.3	95.4	76.2	62.7	73.9	82.3	77.1	73.9
+ Cond-Diff	47.4	62.9	88.4	80.6	97.2	77.4	64.8	79.4	73.9	83.6	75.1
+ Mix-Gen	72.4	60.7	94.0	86.6	98.2	85.2	73.6	85.6	74.0	80.6	78.3

方法	ShipsEar数据集准确率						DeepShip数据集准确率
方法	A类	B类	C类	D类	E类	OA	货船	客船	油轮	拖船	OA
ResNet18	47.4	58.4	85.2	87.5	99.8	77.9	64.8	74.5	74.9	79.1	73.1
+ SpecAugment	57.1	58.4	90.8	85.6	100.0	81.1	68.3	75.3	72.4	82.8	74.3
+ GAN	49.4	67.4	90.8	88.4	93.6	80.6	65.9	77.4	76.5	81.4	75.1
+ Mixup	58.3	64.7	90.2	86.8	95.5	81.4	67.8	78.1	78.2	82.4	75.4
+CutMix	54.8	70.4	91.3	91.1	97.6	81.6	67.1	78.6	77.6	82.2	75.2
+ Mix-Gen	44.9	76.4	89.8	92.6	98.2	81.9	66.7	76.4	76.3	85.0	75.7
VGG16	54.5	74.2	84.9	84.3	92.7	79.0	63.2	74.5	74.1	75.6	71.7
+ SpecAugment	50.6	64.0	87.3	93.1	89.9	80.0	63.7	73.1	76.9	79.0	72.6
+ GAN	55.1	67.4	87.5	93.6	93.5	80.5	65.2	74.8	71.6	78.4	72.2
+ Mixup	56.5	69.8	88.5	92.8	94.0	81.3	63.0	73.0	76.0	78.5	72.6
+CutMix	55.0	71.5	87.0	93.5	96.0	80.9	64.0	73.5	72.5	78.8	72.1
+ Mix-Gen	60.3	58.4	93.7	88.9	96.3	83.0	66.0	70.7	75.8	79.7	72.8
ViT	28.2	24.7	88.7	70.4	91.7	66.7	64.7	67.8	67.8	76.8	68.9
+ SpecAugment	39.1	42.7	96.1	85.6	99.1	77.9	66.7	76.8	62.9	85.2	72.3
+ GAN	54.0	67.1	89.5	93.5	94.1	80.7	66.4	70.9	69.7	80.6	71.5
+ Mixup	58.0	69.5	90.0	92.5	94.0	81.0	66.5	73.0	66.0	82.5	71.9
+CutMix	57.5	69.0	90.0	92.0	93.5	80.5	66.4	72.9	66.1	82.4	71.8
+ Mix-Gen	53.8	75.3	90.8	88.4	86.2	81.3	67.0	73.1	71.3	79.1	72.3
CNN	42.9	68.5	84.2	83.3	95.4	76.2	62.7	73.9	82.3	77.1	73.9
+ SpecAugment	48.7	59.6	93.3	92.1	100.0	76.2	66.0	68.2	75.9	84.4	73.2
+ GAN	53.2	57.3	93.1	98.4	99.1	82.0	64.8	79.4	73.9	83.6	75.7
+ Mixup	74.4	56.2	92.6	71.8	99.1	81.3	68.5	81.0	80.1	80.3	77.4
+CutMix	57.1	48.3	93.0	89.8	98.2	81.6	70.5	82.5	81.5	77.5	77.9
+ Mix-Gen	72.4	60.7	94.0	86.6	98.2	85.2	73.6	85.6	74.0	80.6	78.3

[1]	陈任飞, 陈佳慧, 李忠文, 王思霖. 基于多相机信息协同的水面漂浮物跟踪方法研究[J]. 系统工程与电子技术, 2026, 48(8): 2545-2559.
[2]	韩成杰, 彭聪, 施苏牧, 王宇, 纪江南. 基于多尺度交互融合网络的作动器故障诊断[J]. 系统工程与电子技术, 2026, 48(7): 2319-2332.
[3]	袁胜智, 王少蕾, 李静, 肖任楷, 谭浩天, 赖建伟, 许江涛. 基于小波变换和改进残差神经网络的故障诊断方法[J]. 系统工程与电子技术, 2026, 48(6): 2031-2041.
[4]	陈垣光, 刘晓凯, 肖思瑶, 张顺生, 崔红元, 陈晓莹, 刘莹. 低信噪比下面向时空特征融合的雷达目标智能检测方法[J]. 系统工程与电子技术, 2026, 48(5): 1523-1531.
[5]	杨泽莹, 陈卓, 徐志明, 艾小锋, 吴其华, 王玲. 空间目标ISAR图像部件正向标记方法[J]. 系统工程与电子技术, 2026, 48(4): 1165-1173.
[6]	林炅, 高朋, 洪文兴. 物理信息神经网络在空天领域的应用与展望[J]. 系统工程与电子技术, 2026, 48(4): 1264-1282.
[7]	程鲲, 雷洪涛, 吕志轩. 改进YOLOv8的轻量级无人机跟踪方法[J]. 系统工程与电子技术, 2026, 48(3): 737-750.
[8]	张龙, 朱连宏, 杨波, 雷震, 冯轩铭. MFA-Net：一种面向复杂对抗环境的反舰导弹智能识别多模态自适应融合网络[J]. 系统工程与电子技术, 2026, 48(3): 894-907.
[9]	田弘宇, 葛松虎, 郭宇, 崔中普, 梁潇. 低干噪比下多特征融合的通信干扰识别方法[J]. 系统工程与电子技术, 2026, 48(3): 1061-1071.
[10]	包玉刚, 贾皓翔, 赵旦峰. 基于高阶递归网络的单幅图像去雨滴模型[J]. 系统工程与电子技术, 2026, 48(1): 12-21.
[11]	陈海青, 汪刘应, 刘顾, 王龙, 葛超群, 陈孟州. 合成孔径雷达图像去噪算法研究进展[J]. 系统工程与电子技术, 2026, 48(1): 94-105.
[12]	李开明, 张袁鹏, 罗迎, 代肖楠. 弹道导弹雷达目标识别研究进展[J]. 系统工程与电子技术, 2025, 47(9): 2870-2889.
[13]	季策, 马相宇, 牟晓宇, 赵家毅. TS-GRU-VTA：基于深度学习的车辆信道估计方案[J]. 系统工程与电子技术, 2025, 47(9): 3093-3098.
[14]	闫小伟, 凌冲, 石胜斌. 地表未爆子弹药快速检测系统设计与实现[J]. 系统工程与电子技术, 2025, 47(8): 2639-2645.
[15]	张琬滢, 高由兵, 李泽一, 李鹏飞, 张伟. 基于双路径特征融合网络的背景信号抑制算法[J]. 系统工程与电子技术, 2025, 47(7): 2406-2413.

基于混合扩散模型的水声数据增扩方法

Mixed diffusion model-based underwater acoustic data augmentation method

RichHTML

PDF (PC)

可视化

摘要/Abstract

引用本文

使用本文

图/表 10

参考文献 30

相关文章 15

编辑推荐

Metrics

本文评价