基于小样本条件下水下回波生成方法

王晨宇; 刘礼文; 刘罡; 李博; 杨云川

doi:10.11993/j.issn.2096-3920.202108015

基于小样本条件下水下回波生成方法

doi: 10.11993/j.issn.2096-3920.202108015

中国船舶集团有限公司第705研究所, 陕西西安, 710077

基金项目: 中国博士后科学基金第68批面上项目(2020M683597).

详细信息

通讯作者:
刘礼文(1989-), 男, 博士, 在站博士后, 主要研究方向为人工智能水下目标识别.

中图分类号: TJ630.1; TB71.2
计量
- 文章访问数: 142
- HTML全文浏览量: 50
- PDF下载量: 34
- 被引次数: 0
出版历程
- 收稿日期: 2021-08-30
- 修回日期: 2021-11-09
- 录用日期: 2022-08-15
- 网络出版日期: 2022-09-05

Research on Underwater Echo Generation Method Based on Small Sample

The 705 Research Institute, China State Shipbuilding Corporation Limited, Xi’an 710077, China

摘要

摘要: 在人工智能引领的浪潮下, 将深度学习方法应用在水下目标识别领域已经成为当前研究的热点之一。然而, 在实际科学研究中由于受环境、时间、成本等多方面因素的限制, 水下样本数据的获取变得极其困难, 样本总量的不足导致深度学习模型的训练效果不佳。生成对抗网络作为一种新的人工智能技术, 在数据增强、图像生成等领域具有广泛的应用。然而, 传统结构的生成对抗网络模型对水下回波样本并不适用, 不能直接用于样本数据生成。因此, 针对水下目标数据不足问题, 提出了基于改进型生成对抗网络的小样本条件下水下回波信号生成方法, 结合回波信号的特点, 设计并搭建了基于卷积单元的生成对抗网络模型, 并且利用水池实验测试数据进行了回波信号生成仿真实验。最后在信号波形和幅度概率分布层面验证了生成信号的有效性。实验结果表明, 文中提出的改进型生成对抗网络模型适用于小样本的情况下高度逼真原始回波信号的生成, 为水下目标的主动探测、识别提供了新的思路。
- 水下回波信号 /
- 数据增强 /
- 深度学习 /
- 生成对抗网络
Abstract: Under the tide led by artificial intelligence, the application of deep learning methods in underwater target recognition has gained increasing attention. However, obtaining underwater sample data from actual scientific research is extremely difficult because of environmental constraints, time, cost, and other factors. An insufficient number of samples leads to poor training effects in deep learning models. Generative adversarial networks(GAN), a new artificial intelligence technology, have a wide range of applications in data enhancement, image generation, and other fields. However, the traditional GAN model is unsuitable for underwater echo samples and cannot be used directly for sample data generation. Therefore, this study proposed an underwater echo signal generation method based on an improved GAN for a small sample to address the problem of insufficient underwater target data. Combining the characteristics of the echo signal, the study designed and built a convolution unit-based GAN model. Furthermore, an echo signal generation simulation experiment was carried out using the pool experimental test data. Finally, the effectiveness of the generated signal was verified at the level of the signal waveform and amplitude probability distribution. The experimental results show that the improved GAN model proposed in this study is suitable for generating highly realistic original echo signals in the case of small samples, which provides a new idea for the active detection and recognition of underwater targets.
- underwater echo signal /
- data enhancement /
- deep learning /
- generative adversarial networks

HTML全文

图 1 生成对抗网络结构示意图

Figure 1. Structure diagram of GAN

下载: 全尺寸图片幻灯片

图 2 生成网络详细结构

Figure 2. Detailed structure diagram of generative network

下载: 全尺寸图片幻灯片

图 3 判别网络详细结构

Figure 3. Detailed structure diagram of discriminative network

下载: 全尺寸图片幻灯片

图 4 生成对抗网络训练过程示意图

Figure 4. Schematic diagram of GAN training process

下载: 全尺寸图片幻灯片

图 5 水池实验侧视示意图

Figure 5. Side view of pool experiment

下载: 全尺寸图片幻灯片

图 6 数据处理流程

Figure 6. Data processing flow

下载: 全尺寸图片幻灯片

图 7 相关接收机原理

Figure 7. Principle of correlation receiving machine

下载: 全尺寸图片幻灯片

图 8 处理前后的回波信号

Figure 8. Echo signal before and after processing

下载: 全尺寸图片幻灯片

图 9 训练100回合后生成信号与水池测试信号对比

Figure 9. Comparison between generated signal and pool test signal after 100 epoch of training

下载: 全尺寸图片幻灯片

图 10 训练150回合后生成信号与水池测试信号对比

Figure 10. Comparison between generated signal and pool test signal after 150 epoch of training

下载: 全尺寸图片幻灯片

图 11 训练200回合后生成信号与水池测试信号对比

Figure 11. Comparison between generated signal and real signal after 200 epoch of training

下载: 全尺寸图片幻灯片

图 12 训练250回合后生成信号与水池测试信号对比

Figure 12. Comparison between generated signal and real signal after 250 epoch of training

下载: 全尺寸图片幻灯片

表 1 训练超参数选择

Table 1. Training hyper-parameter selection

优化器参数 Adam参数

L2惩罚系数 0.001
批量大小 4
Leaky ReLU斜率系数 0.2
生成网络学习率 0.000 3
判别网络学习率 0.000 1

下载: 导出CSV

参考文献(22)

[1]	尹韶平, 刘瑞生. 鱼雷总体技术[M]. 北京: 国防工业出版社, 2011.
[2]	陈春玉, 王岚. 反鱼雷技术[M]. 北京: 国防工业出版社, 2006.
[3]	He K, Zhang X, Ren S, et al. Spatial Pyramid Pooling in Deep Convolutional Networks for Visual Recognition[J]. IEEE Transactions on Pattern Analysis & Machine Intelligence, 2014, 37(9): 1904-16.
[4]	Hasan R M, Farshad A, Ali S S. Audio-visual Feature Fusion via Deep Neural Networks for Automatic Speech Recognition[J]. Digital Signal Processing, 2018, 82: 54-63.
[5]	Mclaren M, Lei Y, Scheffer N, et al. Application of Convolutional Neural Networks to Speaker Recognition in Noisy Conditions[C]// Proceedings of the Annual Conference of the International Speech Communication Association. [S.l.]: ISCA, 2014.
[6]	Xu C, Quan W, Song Y, et al. An Optimal Faster-RCNN Algorithm for Intelligent Battlefield Target Recognition[C]// 2020 IEEE International Conference on Artificial Intelligence and Computer Applications (ICAICA). Dalian, China: IEEE, 2020.
[7]	Zhang Y, Marshall I, Wallace B C. Rationale-Augmented Convolutional Neural Networks for Text Classification[EB/OL]. (2016-09-24)[2021-05-03]. https://doi.org/10.48550/arXiv.1605.04469.
[8]	肖璐, 冯西安. 基于K-L变换的水下目标声图像识别方法研究[J]. 计算机仿真, 2013, 30(3): 360-363. doi: 10.3969/j.issn.1006-9348.2013.03.087 Xiao Lu, Feng Xi-an. Research on Acoustical Imaging Recognition Method Based on K-L Transformation[J]. Computer Simulation, 2013, 30(3): 360-363. doi: 10.3969/j.issn.1006-9348.2013.03.087
[9]	代宏伟. 基于人工神经网络的水下探测目标的自动识别[J]. 毕节学院学报, 2011, 29(8): 112-117. Dai Hong-wei. Based on Artificial Neural Networks Automatic Recognition of Underwater Target Detection on Behalf of the Grand[J]. Journal of Bijie University, 2011, 29(8): 112-117.
[10]	朱炜. 基于粒子群的水下图像分割与识别技术研究[D]. 哈尔滨: 哈尔滨工程大学, 2008.
[11]	曾向阳. 智能水中目标识别[M]. 北京: 国防工业出版社, 2016.
[12]	Wang J, Chen Y, Gu Y, et al. Sensory GANs: An Effective Generative Adversarial Framework for Sensor-based Human Activity Recognition[C]//2018 International Joint Conference on Neural Networks. Rio de Janeiro, Brazil: IEEE, 2018.
[13]	佟博, 刘韬, 刘畅. 基于生成对抗网络的轴承失效信号生成研究[J]. 电子科技, 2020, 33(4): 32-38. doi: 10.16180/j.cnki.issn1007-7820.2020.04.006 Tong Bo, Liu Tao, Liu Chang. Research on Bearing Failure Signal Generation Based on Generative Adversarial Networks[J]. Electronic Science and Technology, 2020, 33(4): 32-38. doi: 10.16180/j.cnki.issn1007-7820.2020.04.006
[14]	杨鸿杰, 陈丽, 张君毅. 基于生成对抗网络的数字信号生成技术研究[J]. 电子测量技术, 2020, 43(20): 127-132. doi: 10.19651/j.cnki.emt.2004968 Yang Hong-jie, Chen Li, Zhang Jun-yi. Research on Digital Signal Generation Technology Based on Generative Adversarial Network[J]. Electronic Measurement Technology, 2020, 43(20): 127-132. doi: 10.19651/j.cnki.emt.2004968
[15]	Kataoka Y, Matsubara T, Uehara K. Image Generation Using Generative Adversarial Networks and Attention Mechanism[C]//IEEE/ACIS International Conference on Computer & Information Science. Okayama, Japan: IEEE, 2016.
[16]	Mao X, Wang S, Zheng L, et al. Semantic Invariant Cross-domain Image Generation with Generative Adversarial Networks[J]. Neurocomputing, 2018, 293(7): 55-63.
[17]	Wang K F, Gou C, Duan Y J, et al. Generative Adversarial Networks: The State of the Art and Beyond[J]. Acta AutomaticaSinica, 2017, 43(3): 321-332.
[18]	Zhang C, Liang M, Song X, et al. Generative Adversarial Network for Geological Prediction Based on TBM Operational Data[J]. Mechanical Systems and Signal Processing, 2022, 162(1): 108035.
[19]	Song S, Mukerji T, Hou J. Geological Facies Modeling Based on Progressive Growing of Generative Adversarial Networks(GANs)[J]. Computational Geosciences, 2021, 25(3): 1251-1273. doi: 10.1007/s10596-021-10059-w
[20]	Hershey J R, Olsen P A. Approximating the KullbackLeibler Divergence Between Gaussian Mixture Models[C]// IEEE International Conference on Acoustics. Honolulu, HI, USA: IEEE, 2007.
[21]	刘伯胜, 黄益旺, 陈文剑, 等. 水声学原理[M]. 北京: 科学出版社, 2019.
[22]	邓力, 俞栋. 深度学习方法及应用[M]. 北京: 机械工业出版社, 2015.