摘要
针对典型弹道锥体目标分类需构造、提取人工特征而缺乏通用性及智能性的问题,提出一种利用卷积神经网络(Convolutional Neural Network,CNN)与长短期记忆网络(Long Short-Term Memory,LSTM)相结合的网络模型,对弹道锥体目标的微动时频图实现智能分类的方法。首先,分析弹道锥体目标的微多普勒特征,得出不同微动形式的微多普勒频率;然后,利用AlexNet网络的图像特征提取能力与BiLSTM网络的时序特征提取能力构造AlexNet-BiLSTM网络模型,并通过模拟雷达回波生成的时频图数据集对网络进行训练、调试;最后,仿真结果表明该网络能实现弹道锥体目标的智能微动分类,验证了该网络的有效性及鲁棒性。
Aiming at the problem that the classification of typical ballistic cone targets needs to construct and extract artificial features, but lacks generality and intelligence, a new method of intelligent classification of ballistic cone targets based on micro-motion time-frequency diagram is proposed, which combines Convolutional Neural Network(CNN) with Long Short-Term Memory(LSTM). Firstly, the micro-Doppler characteristics of the ballistic cone target are analyzed, and the micro-Doppler frequencies of different micro-motion forms are obtained;then, the AlexNet-BiLSTM network model is constructed by using the image feature extraction ability of AlexNet network and the temporal feature extraction ability of BiLSTM network, and the network is trained and debugged by using time-frequency diagram data set generated by simulated radar echoes. Finally, the simulation results show that the network can achieve intelligent micro-motion classification of ballistic cone targets, which verifies the effectiveness and robustness of the network.
作者
李江
冯存前
王义哲
许旭光
Li Jiang;Feng Cunqian;Wang Yizhe;Xu Xuguang(Graduate School of Air Force Engineering University,Xi’an,Shaanxi 710051,China;Air and Missile Defense College,Air Force Engineering University,Xi’an,Shaanxi 710051,China;Collaborative Innovation Center of Information Sensing and Understanding,Xi’an,Shaanxi 710077,China)
出处
《信号处理》
CSCD
北大核心
2019年第11期1835-1843,共9页
Journal of Signal Processing
基金
国家自然科学基金(61701528)
关键词
弹道目标
微多普勒
时频分析
深度学习
微动分类
ballistic target
micro-Doppler
time-frequency analysis
deep learning
micro-motion classification
