基于多维特征神经网络集成的有源干扰识别算法  

Active jamming recognition algorithm based on multi⁃dimensional feature neural network integration

作　　者：赵忠臣 刘利民 解辉 韩壮志 荆贺 ZHAO Zhongchen;LIU Limin;XIE Hui;HAN Zhuangzhi;JING He(Department of Electronic and Optics Engineering,Shijiazhuang Campus of Army Engineering University of PLA,Shijiazhuang 050003,China)

机构地区：[1]陆军工程大学石家庄校区电子与光学工程系,河北石家庄050003

出　　处：《现代电子技术》2023年第19期1-7,共7页Modern Electronics Technique

摘　　要：针对在强噪声环境下雷达有源干扰识别准确率不高的问题,提出一种基于功率谱密度、频谱瞬时包络与时频图特征神经网络集成的干扰识别算法。首先从理论推导和方法实现两个角度论述了神经网络集成的基本原理;然后阐述了特征获取方法、网络结构和采用Stacking策略进行网络生成与集成的过程,分析了有源干扰识别的训练、测试结果;最后将该方法与基于时频图的AlexNet检测器、基于功率谱密度序列的LSTM检测器、基于特征融合的双通道检测器进行比较,仿真结果表明,在干噪比(JNR)大于-6 dB时,所提算法对6种有源干扰识别准确概率达到99%以上,具有很好的检测性能。In view of the low recognition accuracy rate of radar active jamming in strong noise environment,a jamming recog⁃nition algorithm based on power spectral density,instantaneous envelope of spectrum and time⁃frequency graph feature neural net⁃work is proposed.The basic principles of neural network integration are discussed in the perspectives of theoretical derivation and method implementation.The feature acquisition method,network structure,and the process of network generation and integra⁃tion with stacking strategy are introduced.The training and testing results of active jamming recognition are analyzed.The pro⁃posed method is compared with the AlexNet detector based on time⁃frequency graph,the LSTM(long short⁃term memory)detector based on power spectral density sequence,and the two⁃channel detector based on feature fusion.The simulation results show that,when JNR is greater than-6 dB,the accuracy probability of identifying six kinds of active interference is over 99%,showing good detection performance.

关 键 词：有源干扰识别 神经网络集成 多维特征 深度自编码器 卷积神经网络 Stacking策略 机器学习 泛化误差 

分 类 号：TN974-34[电子电信—信号与信息处理]