摘要
基于现阶段应用于雷达抗干扰研究中的深度神经网络模型大多是在大量有标签的仿真数据样本上进行多次优化更新,当样本数量受限时,极易出现参数估计偏差大及过拟合问题,文中提出了一种基于Wasserstein生成对抗网络的小样本抗主瓣干扰目标检测方法。该方法首先利用深度神经网络构建了从接收到抗干扰检测的端到端的处理过程,然后采用Wasserstein生成对抗网络学习相应的样本分布实现对回波数据的增广,使得抗干扰检测网络能够通过充分训练获得更具判别性的特征表示,进而实现理想的检测效果。基于小样本数据集的实验结果表明:该方法能够驱动检测网络获得更好的抗干扰和目标检测性能,验证了所提方法的有效性。
At present,most of the deep neural network models used in radar anti-jamming research are optimized and updated on a large number of labeled simulation data samples.When the number of samples is limited,it is easy to cause parameter estimation deviation and overfitting problems,so a small sample mainlobe jamming suppression and target detection method based on Wasserstein generative adversarial network is proposed.In this method,the end-to-end process from receiving to anti-jamming detection is constructed by deep neural network.Then the Wasserstein generative adversarial network is used to learn the sample distribution to realize the augmentation of echo data,so that the anti-jamming detection network can obtain more discriminative feature representation through full training and achieve the ideal detection effect.The results based on small samples′data set show that the proposed method can derive the detection network to obtain better anti-jamming and detection performance,which validate the effectiveness of the proposed method.
作者
陈琳滨
张钰
纠博
刘宏伟
CHEN Linbin;ZHANG Yu;JIU Bo;LIU Hongwei(National Laboratory of Radar Signal Processing,Xidian University,Xi'an 710071,China)
出处
《现代雷达》
CSCD
北大核心
2022年第2期49-57,共9页
Modern Radar
