Page 121 - 应用声学2019年第2期

第 38 卷 第 2 期 Vol. 38, No. 2
2019 年 3 月 Journal of Applied Acoustics March, 2019

⋄ 研究报告 ⋄

水下声目标的梅尔倒谱系数智能分类方法

张少康 1† 田德艳 2

(1 海军潜艇学院 青岛 266000)
(2 青岛海洋科学与技术试点国家实验室 青岛 266000)

摘要 传统水下声目标识别分类方法具有较强的人机交互特性，无法满足未来水下无人平台智能识别分类水
声目标的需求。针对这一问题，提出了一种基于梅尔倒谱系数的水下声目标智能识别分类方法，该方法通过提
取水下声目标梅尔倒谱系数特征，采用长短时记忆网络构建了智能识别分类模型。使用实际水声信号对该方
法进行了验证，结果表明，基于梅尔倒谱系数的水下声目标智能识别分类方法能够在不依赖人工提取特征的
情况下，对目标噪声进行识别分类，具备智能化识别分类能力。
关键词 水下声目标识别分类，梅尔倒谱系数，长短时记忆网络，智能分类
中图法分类号: TP391.4 文献标识码: A 文章编号: 1000-310X(2019)02-0267-06
DOI: 10.11684/j.issn.1000-310X.2019.02.017

Intelligent classification method of Mel frequency cepstrum coefficient for
underwater acoustic targets

ZHANG Shaokang 1 TIAN Deyan 2

(1 Navy Submarine Academy, Qingdao 266000, China)
(2 National Laboratory for Marine Science and Technology, Qingdao 266000, China)

Abstract The traditional methods of underwater target noise recognition have strong human-computer inter-
action characteristics, which can not meet the requirements of intelligent underwater acoustic target recognition
for the future unmanned underwater platform. To solve this problem, an intelligent recognition method of un-
derwater target noise based on Mel frequency cepstrum coefficient (MFCC) is proposed. By extracting the
features of Mel frequency cepstrum coefficient, an intelligent recognition model is constructed by using long
short-term memory network (LSTM). The underwater acoustic signal is used to verify the method. The results
show that the method of underwater target noise intelligent recognition based on Mel frequency cepstrum co-
efficient can identify the target noise without relying on the artificial feature extraction and have an intelligent
recognition ability.
Key words Underwater acoustic targets classification, Mel frequency cepstrum coefficient, Long short-term
memory network, Intelligent classification

2018-09-05 收稿; 2018-12-10 定稿
作者简介: 张少康 (1990- ), 男, 河北石家庄人, 博士研究生, 研究方向: 海洋环境效应技术。
† 通讯作者 E-mail: darth_zhang@163.com