Page 135 - 《应用声学》2021年第1期
P. 135
第 40 卷 第 1 期 Vol. 40, No. 1
2021 年 1 月 Journal of Applied Acoustics January, 2021
⋄ 研究报告 ⋄
非高斯环境下的深度学习脉冲信号去噪与重构 ∗
李 悦 1,2,3 马晓川 1,2† 王 磊 1,2 刘 宇 1,2
(1 中国科学院声学研究所 北京 100190)
(2 中国科学院水下航行器信息技术重点实验室 北京 100190)
(3 中国科学院大学 北京 100049)
摘要：由于实际海洋环境中存在大量的非高斯噪声，一些基于高斯假设的传统去噪方法在实际海洋环境中性
能下降甚至失效。针对非高斯噪声，如 α 稳定分布噪声、非平稳行船噪声下的脉冲信号的去噪与重构，该文提
出一种基于深度学习的方法。去噪模型首先通过学习带噪信号短时傅里叶变换谱与残差谱之间的映射关系以
去除环境噪声，之后对去噪信号的时频谱进行逆变换重构脉冲信号。仿真实验结果表明，深度学习模型在非高
斯噪声环境下脉冲信号的去噪与重构任务中有着良好的表现，在实测样本上也表现出良好的泛化性，体现了
一定的工程应用价值。
关键词：深度学习；非高斯噪声；信号去噪
中图法分类号: O429 文献标识码: A 文章编号: 1000-310X(2021)01-0131-11
DOI: 10.11684/j.issn.1000-310X.2021.01.016
Using deep learning to de-noise and reconstruct pulse signals in
non-Gaussian environment
LI Yue 1,2,3 MA Xiaochuan 1,2 WANG Lei 1,2 LIU Yu 1,2
(1 Institute of Acoustics, Chinese Academy of Sciences, Beijing 100190, China)
(2 Key Laboratory of Information Technology for AUV’s, Chinese Academy of Sciences, Beijing 100190, China)
(3 University of Chinese Academy of Sciences, Beijing 100049, China)
Abstract: Due to the ubiquitous non-Gaussian noise in marine environments, traditional denoising methods
based on Gaussian assumption may degrade or even fail in these circumstances. A deep learning based method
is proposed in this paper, aiming at the denoising and reconstruction problem of pulse signals in non-Gaussian
noise, including α-stable distributed noise and non-stationary ship noise. First, the mapping relationship
between the short-time Fourier transform characteristics of noisy signal and the residual signal is learned for
environmental noise removal. Then, the reconstructed signal is obtained through the inverse short-time Fourier
transform on the denoised time-frequency spectrum. Simulation results show that the proposed method has a
good performance in denoising and reconstruction tasks of pulse signals in non-Gaussian noise circumstances,
as well as good generalization on the test sample in actual situation. Therefore, it shows promising prospect
in engineering application.
Keywords: Deep learning; Non-Gaussian noise; Signal denoising
2020-04-28 收稿; 2020-07-15 定稿
中国科学院先导 A 项目轻巡航器课题
∗
作者简介: 李悦 (1992– ), 女, 陕西合阳人, 博士研究生, 研究方向: 信号与信息处理。
通信作者 E-mail: maxc@mail.ioa.ac.cn
†
