Page 79 - 《应用声学》2021年第5期
P. 79

第 40 卷 第 5 期                                                                       Vol. 40, No. 5
             2021 年 9 月                          Journal of Applied Acoustics                 September, 2021

             ⋄ 研究报告 ⋄



                     基于广义回归神经网络的强干扰下垂直阵


                                          目标距离估计方法                           ∗





                                        姚琦海 汪 勇            †   黎佳艺 杨益新



                                               (西北工业大学航海学院       西安   710072)

                摘要:以声压场采样协方差矩阵为特征,基于广义回归神经网络研究在强干扰下的水下声源测距问题,该文
                提出了优化扩展因子的方法以提高神经网络估计性能。使用仅有一个网络参数的广义回归神经网络,使用
                SWellEX-96 实验 S59 航次的垂直阵数据,比较了以传统匹配场处理为代表的模型驱动方法和以卷积神经网
                络、广义回归神经网络为代表的数据驱动方法在强干扰下的水下目标距离估计性能。结果表明,基于优化扩展
                因子的广义回归神经网络在强干扰下可以有效实现距离估计。
                关键词:距离估计;强干扰;垂直阵;广义回归神经网络
                中图法分类号: TB533           文献标识码: A          文章编号: 1000-310X(2021)05-0723-08
                DOI: 10.11684/j.issn.1000-310X.2021.05.010





                        Source range estimation method of vertical array under strong

                                            interference based on GRNN



                                    YAO Qihai    WANG Yong      LI Jiayi  YANG Yixin

                      (School of Marine Science and Technology, Northwestern Polytechnical University, Xi’an 710072, China)

                 Abstract: The sample covariance matrix of sound pressure field is made as the feature. The research on
                 underwater sound source ranging under strong interference based on generalized regression neural network
                 (GRNN), which has only one network parameter, extension factor. It proposes a method of optimizing the
                 extension factor to improve the estimation performance of neural network. The research uses the vector line
                 array (VLA) data from event S59 of the SWellEx-96 experiment, comparing the range estimation performance
                 of underwater targets under strong interference of model-driven traditional matched field processing, data-
                 driven convolutional neural networks (CNN) and GRNN. The results show that GRNN based on the optimized
                 extension factor can effectively realize the estimation of range under strong interference.
                 Keywords: Range estimation; Strong interference; Vector line array; Generalized regression neural network





             2020-11-13 收稿; 2021-01-24 定稿
             国家自然科学基金项目 (61971353)
             ∗
             作者简介: 姚琦海 (1997– ), 男, 陕西西安人, 硕士研究生, 研究方向: 信号与信息处理。
             † 通信作者 E-mail: yongwang@nwpu.edu.cn
   74   75   76   77   78   79   80   81   82   83   84