Generation Method of Underwater Samples Based on a Generative Adversarial Network
-
摘要: 神经网络技术已成为水下高速航行器目标检测的应用趋势, 该技术需要大量的训练样本以保证训练结果的准确性。生成对抗网络(GAN)作为解决训练样本稀少问题的重要方法, 被广泛应用在各个领域。文中针对水下样本特点对经典GAN模型进行改进, 提出一种基于GAN的水下样本生成方法, 以达到扩增训练样本的目的。首先构建适用于水下样本的GAN模型, 然后以实航试验数据训练模型并优化参数, 最后用该模型进行样本生成仿真并验证生成结果的有效性。仿真结果表明, 生成样本与试验样本吻合较好, 可实现试验样本的数据增强。该方法将有助于解决水下数据样本稀少问题, 为进一步应用神经网络提高水下高速航行器目标检测的效率和准确率提供参考。Abstract: Neural network technology has become an application trend in the area of target detection of undersea high-speed vehicles, but the technology requires numerous training samples to ensure the accuracy of training results. As an important method used to solve the problem of sparse training samples, generative adversarial networks (GANs) are widely used in various fields. This study improves the classic GAN model based on the characteristics of underwater samples and proposes a generation method for underwater samples based on GAN for the purpose of augmenting training samples. First, a GAN model suitable for underwater samples is constructed, and the actual sea trial data are used to train the model and optimize the parameters. Finally, the model is used to simulate the generation of samples and verify the effectiveness of the results. Simulation results show that the generated and test samples are in good agreement and that data augmentation of the test samples can be realized. This method helps solve the problem of sparse underwater data samples and provides a reference for further application of neural networks to improve the efficiency and accuracy of undersea high-speed vehicle target detection
-
[1] 李志舜. 鱼雷自导信号与信息处理[M]. 西安: 西北工业大学出版社, 2004. [2] 周德善. 鱼雷自导技术[M]. 北京: 国防工业出版社, 2009. [3] 宋达. 基于深度学习的水下目标识别方法研究[D]. 成都: 电子科技大学, 2018. [4] 赵增科. 基于深度学习的水下目标识别[D]. 哈尔滨: 哈尔滨工程大学, 2017. [5] 凡志邈, 李海林, 夏伟杰, 等. 基于深度学习的成像声呐水下目标的检测与分类[C]//中国声学学会水声学分会2019年学术会议论文集. 南京: 中国声学学会水声学分会, 2019. [6] Goodfellow I J, Pougetabadie J, Mirza M, et al. Generative Adversarial Nets[J]. Neural Information Processing Systems, 2014, 3: 2672-2680. [7] Schmidhuber J. Deep Learning in Neural Networks: An Overview[J]. Neural Networks, 2015, 61: 85-117. [8] Bengio Y, Laufer E, Alain G, et al. Deep Generative Sto-chastic Networks Trainable by Backprop[J]. International Conference on Machine Learning, 2014, 32(2): 226-234. [9] 张凯, 陈亚军, 张俊. 生成对抗网络在医学小样本数据中的应用[J]. 内江师范学院学报, 2020, 35(4): 57-60.Zhang Kai, Chen Ya-jun, Zhang Jun. Applications of Generative Adversarial Nets in Medical Small Sample Data[J]. Journal of Neijiang Normal University, 2020, 35(4): 57-60. [10] 徐希岩. 基于深度学习的小样本图像分类研究[D]. 哈尔滨: 东北林业大学, 2018. [11] 李秋玮. 基于条件生成对抗网络和超限学习机的小样本数据处理方法研究[D]. 镇江: 江苏大学, 2019. [12] 高强, 姜忠昊. 基于GAN等效模型的小样本库扩增研究[J]. 电测与仪表, 2019, 56(6): 76-81.Gao Qiang, Jiang Zhong-hao. Amplification of Small Sample Library Based on GAN Equivalent Model[J]. Electrical Measurement & Instrumentation, 2019, 56(6): 76-81. [13] 田娟, 李英祥, 李彤岩. 激活函数在卷积神经网络中的对比研究[J]. 计算机系统应用, 2018, 27(7): 45-51.Tian Juan, Li Ying-xiang, Li Tong-yan. Contrastive Study of Activation Function in Convolutional Neural Network[J]. Computer Systems & Applications, 2018, 27(7): 45-51. [14] Ioffe S, Szegedy C. Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift[EB/OL]. arXiv, (2015-03-02)[2020-05-17]. https://arxiv.org/abs/1502.03167. [15] 王坤峰, 苟超, 段艳杰, 等. 生成式对抗网络GAN的研究进展与展望[J]. 自动化学报, 2017, 43(3): 321-332.Wang Kun-feng, Gou Chao, Duan Yan-jie, et al. Generative Adversarial Networks: The State of the Art and Beyond[J]. Acta Automatica Sinica, 2017, 43(3): 321-332. [16] Kingma D P, Ba J. Adam: A Method for Stochastic Optimization[EB/OL]. arXiv, (2014-12-22)[2020-05-27].https://arxiv.org/abs/1412.6980.
点击查看大图
计量
- 文章访问数: 186
- HTML全文浏览量: 0
- PDF下载量: 134
- 被引次数: 0