摘要
针对传统方法在欺骗干扰特征提取时需要依赖人工经验的缺点,提出了基于栈式稀疏自编码器(Stacked Sparse Autoencoder)的有源欺骗干扰识别算法。该算法对干扰下的雷达接收信号进行时频分析,对时频特征进行降维,利用无标签样本对特征提取模型进行预训练,再通过少量有标签样本进行监督精校。最后利用soft max分类器完成有源干扰的识别。仿真实验证明,该方法有较高的识别率,特别是该方法受信噪比影响较少,说明了深度学习方法应用于雷达欺骗干扰信号分类识别领域的可行性。相较于其他文献方法,该算法拥有更好的实验效果,证明了该方法的优越性。
Aimed at the deficiency of traditional technique of radar active deception feature extraction which heavily rely on artificial experience,a recognition algorithm was proposed based on stacked sparse auto-encoder.In this method,spwvd distribution of received radar signal under jamming was given,and dimensionality reduction was implemented with a series of image processing methods.In the phase of pre-training,stacked sparse auto-encoder model was trained with unlabeled samples by greedy layer-wise training.On this basis,network parameters were fine-tuned with label information.Finally,the soft max classifier was used to recognize the active jamming.The simulation results showed that this method had high recognition rate,especially the influence of SNR on this method was less and the feasibility of applying the deep learning method to the classification and recognition of radar deception jamming signal.Compared with other literature methods,the algorithm had better experimental results,and the superiority of the method was proved.
作者
阮怀林
杨兴宇
RUAN Huailin;YANG Xingyu(Electronic Countermeasure Institute of National University of Defense Technology,Hefei 230037,China)
出处
《探测与控制学报》
CSCD
北大核心
2018年第4期62-67,共6页
Journal of Detection & Control
关键词
欺骗干扰
干扰识别
时频分析
深度学习
栈式稀疏自编码器
deception jamming
jamming recognition
time-frequency distribution
deep learning
stacked sparse autoencoder