基于深度学习的弹道目标智能分类  

Intelligent classification of ballistic targets based on deep learning

作　　者：李江 冯存前[1,2] 王义哲 贺思三 LI Jiang;FENG Cunqian;WANG Yizhe;HE Sisan(Air and Missile Defense College,Air Force Engineering University,Xi’an 710051,China;Collaborative Innovation Center of Information Sensing and Understanding,Xi’an 710077,China)

机构地区：[1]空军工程大学防空反导学院,陕西西安710051 [2]信息感知技术协同创新中心,陕西西安710077

出　　处：《系统工程与电子技术》2020年第6期1226-1234,共9页Systems Engineering and Electronics

基　　金：国家自然科学基金(61701526,61701528)资助课题。

摘　　要：针对弹道目标微动分类前需平动补偿及典型雷达散射截面积(radar cross-section,RCS)序列分类需构造人工特征的问题,提出利用弹道目标微动特性和RCS相结合的弹道目标智能分类算法。首先,建立弹道目标运动模型并分析得到方位角和俯仰角,从而获取RCS序列,在此基础上利用小波变换得到时频图并构建数据集;然后,通过卷积神经网络(convolutional neural network,CNN)提取时频图像特征序列并与RCS序列融合成高维特征;最后,利用具有容错能力的双向长短期记忆网络充分学习序列之间的相关性以实现目标分类。仿真结果表明,该算法比卷积神经网络和支持向量机的分类精度分别提高5%和2%以上,分类速度比卷积神经网络和双向长短期记忆网络分别提高1.5倍和2.5倍,实现了更高精度的快速智能分类。Aiming at the problems of translational compensation before micro-motion classification of ballistic targets and the need to construct artificial features for typical radar cross-section(RCS)sequence classification,an intelligent classification method of ballistic targets based on micro-motion characteristics of ballistic targets and RCS is proposed.Firstly,the ballistic targets motion model is established and the azimuth and elevation angles are analyzed to obtain the RCS sequence.On this basis,the time-frequency diagram is obtained by using wavelet transform to construct the data set.Then,the time-frequency diagram feature sequence is extracted by convolutional neural network(CNN)and fused with the RCS sequence to form high-dimensional features.Finally,the bidirectional long short-term memory network with fault tolerance is used to fully learn the correlation between sequences to achieve target classification.The simulation results show that the classification accuracy of the proposed algorithm is 5% and 2% higher than that of CNN and support vector machines,and the classification speed is 1.5 and 2.5 times faster than that of CNN and bidirectional long short-term memory networks,respectively.The algorithm achieves faster intelligent classification with higher accuracy.

关 键 词：深度学习 弹道目标 智能分类 雷达散射截面积 小波变换 

分 类 号：TN957[电子电信—信号与信息处理]