|
肖海霞(1987—),男,硕士,研究方向为水声信号处理与仿真。 |
收稿日期: 2025-09-01
修回日期: 2025-09-28
网络出版日期: 2026-01-23
基金资助
*海军航空大学托举基金(H2202402002)
Underwater acoustic target signal enhancement algorithm optimized by feature preservation and noise update
Received date: 2025-09-01
Revised date: 2025-09-28
Online published: 2026-01-23
水声目标信号的特征交叠及海洋水声场的多变,使得非负矩阵分解(Nonnegative Matrix Factorization,NMF)应用于水声目标信号增强时效果不理想。为此,结合水声目标信号特征对经典NMF算法进行适应性改进,提出基于特征保持的改进NMF水声信号增强算法。算法首先对NMF特征基矩阵进行β-散度约束和相似检测去冗余,以尺寸不变特性优化NMF特征,同时避免因基向量相似带来的系数分散造成基向量丢失;然后以声呐接收到的实时环境噪声改进NMF噪声基向量的适配性,实现水声目标信号的降噪增强。实验结果表明,相比于经典NMF、流形约束NMF等算法应用于水声目标信号增强,文中方法取得更优的信号增强效果。
肖海霞 , 崔双月 , 李大卫 , 孙明明 , 刘贤忠 , 杨真鑫 . 基于特征保持与噪声更新的水声信号增强算法*[J]. 指挥控制与仿真, 2026 , 48(1) : 55 -59 . DOI: 10.3969/j.issn.1673-3819.2026.01.007
The enhancement effect of the classic Nonnegative Matrix Factorization (NMF) applied to underwater acoustic target signal is unsatisfactory for the feature overlap of underwater acoustic target signal and the variability of ocean underwater acoustic field. And so, an improved NMF underwater acoustic signal enhancement algorithm based on feature preservation is proposed by adaptively improving the classic NMF algorithm using the characteristics of underwater acoustic target signals. The β-divergence constraints and similarity detection is first applied to the NMF feature basis matrix to eliminate redundancy, optimized NMF features with size invariant characteristics, while avoiding the loss of basis vectors due to coefficient dispersion caused by similar basis vectors. And then, the real-time environmental noise received by the sonar is used to improve the adaptability of the NMF noise basis vector, achieving noise reduction and enhancement of the underwater acoustic target signal. The experimental results show that, compared to the classical NMF, the manifold constrained NMF that is used to the underwater acoustic signal enhancement, the proposed method achieved the better signal enhancement effect.
| [1] |
|
| [2] |
张石磊, 简志华, 孙闽红, 等. 采用联合字典优化的噪声鲁棒性语音转换算法[J]. 声学学报, 2019, 44(6): 1 074-1 082.
|
| [3] |
施晶, 朱明, 沈华, 等. 基于VMD和小波阈值的水听器信号去噪方法[J]. 传感技术学报, 2020, 33(7): 1 003-1 012.
|
| [4] |
|
| [5] |
姚琨, 杨吉斌, 张雄伟, 等. 基于多分辨率时频特征融合的声学场景分类[J]. 声学技术, 2020, 39(4): 494-500.
|
| [6] |
|
| [7] |
李煦, 王子腾, 王晓飞, 等. 采用性别相关的深度神经网络及非负矩阵分解模型用于单通道语音增强[J]. 声学学报, 2019, 44(2): 221-230.
|
| [8] |
|
| [9] |
王梦阳, 王华庆, 董方, 等. 基于EVMD-LNMF的复合故障信号分离方法[J]. 振动与冲击, 2019, 38(16): 146-152.
|
| [10] |
陈善学, 刘荣华. 基于子空间结构正则化的L21非负矩阵分解高光谱解混[J]. 电子与信息学报, 2022, 44(5): 1 704-1 713.
|
| [11] |
王瀛, 何欣, 左方. 基于最大整体包容度约束非负矩阵分解的高光谱遥感图像混合像元分析算法[J]. 光子学报, 2018, 47(3): 142-150.
|
| [12] |
生雪莉, 穆梦飞, 殷敬伟, 等. 基于稀疏分解的水下运动目标多普勒频移估计方法[J]. 哈尔滨工程大学学报, 2020, 41(10): 1 429-1 435.
|
| [13] |
黄卫春, 赵杨, 熊李艳. 用于独立特征学习的稀疏非负矩阵分解算法[J]. 计算机应用研究, 2020, 37(4): 986-989.
|
| [14] |
|
| [15] |
邱飞岳, 陈博文, 陈铁明, 等. 稀疏诱导流形正则化凸非负矩阵分解算法[J]. 通信学报, 2020, 41(5): 84-95.
|
| [16] |
王华庆, 王梦阳, 宋浏阳, 等. 双约束非负矩阵分解的复合故障信号分离方法[J]. 振动工程学报, 2020, 33(3): 590-596.
|
| [17] |
|
| [18] |
李登刚, 陈香香, 李华丽, 等. 基于超像素的流形正则化稀疏约束NMF混合像元分解算法[J]. 计算机应用, 2019, 39(10): 3 100-3 106.
|
| [19] |
许铭, 王冬霞, 周城旭, 等. 改进的Kullback-Leibler 复非负矩阵分解语音增强算法[J]. 声学技术, 2019, 38(5): 560-567
|
| [20] |
刘贤忠, 吴明辉, 郑晓庆, 等. 改进NMF优化的水声目标信号增强算法[J]. 电光与控制, 2021, 28(9):6-9.
|
/
| 〈 |
|
〉 |