﻿ 基于特征融合的舰船目标识别方法
 舰船科学技术  2022, Vol. 44 Issue (1): 146-149    DOI: 10.3404/j.issn.1672-7649.2022.01.028 PDF

Ship target recognition method based on Feature Fusion
WANG Ying, HOU Peng, WU Di, WANG Wen-guan, LI Guang-yuan
Dalian Scientific Test and Control Technology Institute, Dalian 116013, China
Abstract: The traditional passive target recognition mainly depends on the sonarman. With the rapid development of artificial intelligence, underwater target intelligent recognition becomes the future development trend. To solve this problem, according to the characteristics of ship radiated noise, a method of ship target recognition based on feature fusion is proposed. Based on the Mel frequency cepstrum coefficient feature and spectral correlation density function feature,a feature fusion model is constructed by using convolution neural network.The measured data of four underwater acoustic signal is verified. The result shows the decision level fusion algorithm can improve the recognition accuracy.
Key words: ship radiation noise     Mel frequency cepstrum coefficient     spectral correlation density function     feature fusion     convolution neural network
0 引　言

1 舰船辐射噪声特征提取 1.1 梅尔倒谱特征提取

 ${f_{Mel}} = 2\;595 \times \lg \left( {1 + f/700} \right) ，$ (1)

 图 1 梅尔倒谱系数特征提取过程 Fig. 1 Feature extraction process of Mel cepstrum coefficient

1）预处理

2）计算能量谱

 $E\left( {i,k} \right) = {\left| {X\left( {i,k} \right)} \right|^2} = {\left| {FFT\left[ {{x_i}(m)} \right]} \right|^2}，$ (2)

3）Mel滤波

 $S(i,m) = \sum\nolimits_{k = 0}^{N - 1} {E(i,k){H_m}(k)}，0 \leqslant m < M 。$ (3)

4）计算DCT倒谱

 $mfcc(i,n) = \sqrt {\frac{2}{M}\sum\nolimits_{m = 0}^{M - 1} {\log \left[ {S(i,m)} \right]} \cos \left[ {\frac{{\text{π} n(2m - 1)}}{{2M}}} \right]} 。$ (4)

1.2 谱相关函数特征提取

 ${R_x}\left( {\tau ,\alpha } \right) = \left\langle {x{{\left( {t - \frac{\tau }{2}} \right)}^ * }x\left( {t + \frac{\tau }{2}} \right){e^{ - j2\text{π} \alpha t}}} \right\rangle ，$ (5)

 ${S_x}\left( {f,\alpha } \right) = \int_{ - \infty }^\infty {{R_x}\left( {\tau ,\alpha } \right)} {e^{ - j2\text{π} ft}}{\rm{d}}\tau 。$ (6)

2 基于特征融合的识别网络构建 2.1 卷积神经网络

 图 2 卷积神经网络结构 Fig. 2 Structure of convolution neural network
2.2 特征融合

 图 3 基于特征层融合的舰船目标识别 Fig. 3 Ship target recognition based on feature layer fusion

 图 4 基于决策层融合的舰船目标识别 Fig. 4 Ship target recognition based on decision level fusion

3 试验验证

1）输入层

2）卷积层1

8个卷积核、卷积核尺寸为2×2，步长为1；

3）池化层1

4）卷积层2

16个卷积核、卷积核尺寸为2×2，步长为1；

5）池化层2

6）全连接层

7）输出层

4 结　语

 [1] 程锦盛, 杜选民, 等. 采用深度学习方法的水下目标听觉特征提取与识别技术研究[J]. 声学技术, 2018, 37(6): 167-168. CHENG Jin-sheng, DU Xuan-min, et al. Research on extraction and recognition technique to auditory features of underwater target based on deep learning[J]. Technical Acoustics, 2018, 37(6): 167-168. [2] 曾赛, 杜选民. 水下目标多模态深度学习分类识别研究[J]. 应用声学, 2019, 38(4): 589-594. ZENG Sai, DU Xuanmin. Multimodal underwater target recognition method based on deep learning[J]. Journal of Applied Acoustics, 2019, 38(4): 589-594. DOI:10.11684/j.issn.1000-310X.2019.04.016 [3] 陈雪峰, 张中戈, 等. 基于线谱和DEMON谱的水声目标分类[J]. 指挥信息系统与技术, 2019, 10(4): 61-65. CHEN Xuefeng, ZHANG Zhongge, et al. Classification for underwater acoustic targets based on line spectrum and DEMON spectrum[J]. Command Information System and Technology, 2019, 10(4): 61-65. [4] 韩纪庆, 张磊, 等. 语音信号处理[M]. 北京: 清华大学出版社, 2019. [5] 陈进, 董广明. 机械故障特征提取的循环平稳理论及方法[M]. 上海: 上海交通大学出版社, 2013. [6] 魏秀参. 解析深度学习卷积神经网络原理与视觉实践[M]. 北京: 电子工业出版社, 2018.