在声呐、雷达等设备的目标探测中，声源方位估计是需要解决的关键问题之一。针对水下传感器阵列接收信号的波达方向角（DOA）估计算法中，传统的BP神经网络算法会因网络参数不合理和层数过多导致过拟合的问题，以往通过粒子群算法（PSO）进行优化后，网络仍容易过早结束训练而导致性能不佳。为此，本文提出一种基于变分模态分解结合粒子群算法优化后的BP神经网络算法。首先对目标回波信号进行可变模态分解，对分解得到的各分量进行时频分析后叠加的谱图特征作为经粒子群算法优化后的BP神经网络算法的输入进行训练测试，以此来提高阵元接收目标回波的DOA估计精度。仿真实验结果表明，结合变分模态分解及粒子群算法优化的BP神经网络具有更好的识别效果和泛化能力，提高了DOA的估计精度。
In target detection of sonar, radar and other equipment, the estimation of sound source azimuth is one of the key problems to be solved.In the DOA estimation algorithm of received underwater sensor array signal, the triditional BP neural network algorithm will lead to the problem of over fitting due to unreasonable network parameters and too many layers. In the past, after optimization by particle swarm optimization (PSO), the network is still easy to end training too early, resulting in poor performance. Therefore, this paper proposes a BP neural network algorithm based on the combination of the variational mode decomposition and particle swarm optimization. Firstly, the echo signal of the target is decomposed by the variable mode, and the spectral characteristics of the decomposed components after the time-frequency analysis are used as the input of the BP neural network algorithm optimized by the particle swarm optimization for training and testing, so as to improve the receiving target echo of the array’s DOA estimation accuracy. The simulation results show that the BP neural network combined with the variational mode decomposition and particle swarm optimization has better recognition effect and generalization ability, and improves the estimation accuracy of DOA.
2021,43(7): 122-126 收稿日期：2020-07-29
DOI：10.3404/j.issn.1672-7649.2021.07.025
分类号：TB566
基金项目：国家自然基金资助项目(61540024)；国家自然基金资助项目(61901079)
作者简介：杨阳(1986-),女,博士,讲师,研究领域为水下目标识别
参考文献：
[1] 杜召平, 陈刚, 王达. 国外声呐技术发展综述[J]. 舰船科学技术, 2019, 41(1):149-155
DU Zhao-ping, CHEN Gang, WANG Da. Overview of sonar technology development abroad[J]. Ship Science and Technology, 2019, 41(1):149-155
[2] YANG Jin-sheng, XIANG Yang, CHEN Wei-gang, et al. Computationally efficient 2-DDOA estimation for non-uniform two-L-shaped array[J]. The Journal of China Universities of Posts and Telecommunications, 2018, 25(6):81-89
[3] BURG J P. Maximum entropy spectral analysis[C]//Proc. of the 37th meeting of the Annual Int. SEG Meeting, Oklahoma City, OK, 1967.
[4] CAPON J. High-resolution frequency-wavenumber spectrum analysis[J]. Proceedings of the IEEE, 1969, 57(8):1408-1418
[5] SCHMIDT R, SCHMIDT R O. Multiple emitter location and signal parameters estimation[J]. IEEE Transactions on Antennas & Propagation, 1986, 34(3):276-280
[6] ROY R, KAILATH T. ESPRIT-estimation of signal parameters via rotational invariance techniques[J]. IEEE Transactions on Acoustics, Speech, and Signal Processing, 1989, 37(7):984-995
[7] El ZOOGHBY A H, CHRISTODOULOU C G. A neural network-based smart antenna for multiple source tracking[J]. IEEE Transactions on Antennas and Propagation, 2000, 48(5):P. 768-776
[8] 陈荆花, 易辉跃, 周希朗. 基于特征矢量输入的神经网络测向方法[J]. 上海交通大学学报, 2003(3):373-375+379
[9] 张旻, 李鹏飞. 基于互相关函数相角特征的RBF神经网络来波方位估计[J]. 电子与信息学报, 2009, 31(12):2926-2930
[10] DRAGOMIRETSKIY K, ZOSSO D. Variationalmode decomposition[J]. IEEE Transactions on Signal Processing, 2014, 62(3):531-544
[11] 张炜, 程锦房, 蒋伟涛. 基于WVD-MUISC算法的单矢量水听器DOA估计[J]. 武汉理工大学学报(交通科学与工程版), 2013, 37(4):801-804
[12] 李光远, 侯朋, 程广福. 基于粒子群算法的水下多元线阵阵形有源校正方法[J]. 舰船科学技术, 2017, 39(8):151-155
LI Guang-yuan, HOU Peng, CHENG Guang-fu. Array shape calibration of underwater linearraywith multiple sources based on particle swarm optimization[J]. Ship Science and Technology, 2017, 39(8):151-155
