摘要
随着电磁环境复杂度的不断提升及电子侦察体制的日益复杂,高脉冲丢失率下的复杂雷达脉冲信号分选成为制约电子侦察效能的瓶颈问题。针对实际应用中侦收脉冲数据丢失率高、到达时间抖动、信号体制复杂等问题,提出一种基于长短期记忆网络的变体JANET网络的雷达信号分选方法。该分选算法通过仅增加遗忘门结构,即可有效挖掘脉冲时间序列上下文的特性,实现高丢失率辐射源脉冲的有效分选,同时解决了循环神经网络长序列依赖问题,能够实现脉冲的准实时在线分选,满足工程应用中信号分选准确度及实时性的要求。
Radar signal deinterleaving process is a method of classifying intensive pulse streams.The performance of signal classifiers requires to be improved when being confronted with the large amount of data and mode-switch emitters.Recurrent neural network is appropriate as a classifier for pulse streams.However it is weak of long-term dependencies.The forget gate which is a custom function in JANET overcomes the problem.In this paper,JANET is introduced as a classifier for mining the long-term temporal patterns,and the result proves the breathtaking performance of the proposed method.
作者
姜在阳
孙思月
李华旺
梁广
JIANG Zaiyang;SUN Siyue;LI Huawang;LIANG Guang(Shanghai Institute of Microsystem and Information Technology,Chinese Academy of Sciences,Shanghai 200050,China;Innovation Academy for Microsatellites of Chinese Academy of Sciences,Shanghai 201203,China;University of Chinese Academy of Sciences,Beijing 100049,China)
出处
《中国科学院大学学报(中英文)》
CSCD
北大核心
2021年第6期825-831,共7页
Journal of University of Chinese Academy of Sciences
基金
上海市启明星计划(18QA1404000)和中国科学院青年创新促进会资助。