Page 141 - 《应用声学）》2023年第5期

第 42 卷 第 5 期 Vol. 42, No. 5
2023 年 9 月 Journal of Applied Acoustics September, 2023

⋄ 研究报告 ⋄

基于时延神经网络模型的舰船辐射噪声目标识别 ∗

赵乾坤 刘 峰 † 梁秀兵 汪 涛 宋永强

(军事科学院国防科技创新研究院 北京 100071)

摘要：水声目标被动识别是水声信号处理领域的研究热点之一。深度学习作为一种数据驱动方法，以其对非
线性系统的良好拟合能力，为水声目标识别领域提供了新思路。该文采用一种基于时延神经网络模型的舰船
辐射噪声目标识别方法，该方法利用目标的短时平稳特性和长时关联特性对目标的声纹特征进行建模，使用
梅尔谱图提取目标信号的初级特征，再通过融合注意力机制和时延神经网络的深度学习模型实现高级特性提
取，最后利用余弦相似度实现不同目标的类别划分。该方法在 ShipsEar 数据集和自行采集的数据进行测试验
证，目标识别准确率分别达到 79.2% 和 73.9%，可证明该方法的有效性。
关键词：水声目标识别；舰船辐射噪声；时延神经网络；注意力机制
中图法分类号: TB566 文献标识码: A 文章编号: 1000-310X(2023)05-1033-09
DOI: 10.11684/j.issn.1000-310X.2023.05.017

Target recognition of ship radiated noise based on time delay neural

network model

ZHAO Qiankun LIU Feng LIANG Xiubing WANG Tao SONG Yongqiang

(National Innovation Lnstitute of Defense Technology, Academy of Military Sciences, Beijing 100071, China)

Abstract: Passive recognition of underwater acoustic targets is one of the research hotspots in the field
of underwater acoustic signal processing. As a data-driven method, deep learning provides a new idea for
underwater acoustic target recognition because of its good fitting ability to nonlinear systems. This article
adopts a ship radiated noise target recognition method based on the time delay neural network (TDNN)
model. This method uses the short-term stationary and long-term correlation characteristics of the target to
model the voiceprint features of the target. The Mel spectrum is used to extract the primary features of the
target signal, and then advanced feature extraction is achieved through a deep learning model that integrates
attention mechanism and time-delay neural network. Finally, cosine similarity is used to achieve classification
of different targets. This method was tested and validated on the ShipsEar dataset and self collected data, and
the target recognition accuracy reached 79.2% and 73.9%, respectively, demonstrating the effectiveness of the
proposed method.
Keywords: Underwater acoustic target recognition; Ship radiated noise; Time delay neural network;
Attention mechanism

2023-06-13 收稿; 2023-07-28 定稿
国家自然科学基金项目 (62201608)
∗
作者简介: 赵乾坤 (1992– ), 男, 吉林白山人, 硕士研究生, 研究方向: 水声信号处理。
† 通信作者 E-mail: liufeng_cv@126.com