摘要
针对现有算法不具备识别出未在训练过程中出现的新类别目标的能力问题,提出了针对逆合成孔径成像雷达(inverse synthetic aperture imaging radar,ISAR)隐身目标零样本学习识别方法。首先,基于飞行类动态目标三维网络化物理模型,使用FEKO电磁场仿真软件进行编程,实现ISAR可见的源目标图像数据生成;然后,在此基础上形成不同飞机类目标细节属性的文本语义特征表达。所提出的新类型算法网络模型采用两个变分自编码器,分别进行了图像和语义的特征生成,从而让网络学习到模态不变的特征表达。使用可见类别数据训练网络,并获得能够凭借语义信息生成图像特征的模型。训练识别采用该学习模型不可见的未知新类别目标,从不可见未知的新类别文本语义生成不可见未知的新目标图像特征信息,支撑了不可见未知的新目标识别,统计未知的新类别识别正确率为75%。
For the zero-shot learning and recognition problem of stealth targets for inverse synthetic aperture imaging radar(ISAR),the existed algorithms cannot recognize the new category of targets that have not appeared in the training process.A target learning and recognition method with zero-shot learning(ZSL)is proposed for solving the problem above.Firstly,based on three dimensional grid physical model of the aircraft-class dynamic target,the FEKO electromagnetic field simulation software programming is used to realize the ISAR-visible source target image data generation.Secondly,on this basis,the text semantic feature expression of detail attributes for different aircraft-class targets is formed.In the novel-type proposed algorithm’s network model,two variation autoencoders are used to generate image and semantic features respectively,so that the network can learn the modal’s invariant feature expression.The network is trained with visible category data to obtain the model that can generate image features based on semantic information.The training recognition process adopts the targets of the new category which are unknown and invisible for the proposed learning model.The unknown and invisible image feature information of the new target is generated from the invisible and unknown text semantic of the new category,which supports the recognition for the new targets which are unknown and invisible.The accuracy rate of the recognition for the new category targets is up to 75%according to the relevant statistic result,which means an important military and strategic significance.
作者
周春花
魏维伟
张学成
郑鑫
程冕之
ZHOU Chunhua;WEI Weiwei;ZHANG Xuecheng;ZHENG Xin;CHENG Mianzhi(Shanghai Radio Equipment Research Institute,Shanghai 201109,China;Shanghai Engineering Research Center of Target Identification and Environment Perception,Shanghai 201109,China;Traffic Perception Radar Technology Research&Development Center of China Aerospace Science and Technology Corporation,Shanghai 201109,China;The 3rd Military Representative Office of the Ministry of Army Equipment,Shanghai 200031,China;Shanghai Academy of Spaceflight Technology,Shanghai 201109,China)
出处
《系统工程与电子技术》
EI
CSCD
北大核心
2023年第10期3116-3121,共6页
Systems Engineering and Electronics
基金
上海市自然科学基金(19ZR1454000)资助课题。
关键词
逆合成孔径成像雷达
零样本学习
隐身目标
inverse synthetic aperture imaging radar(ISAR)
zero-shot learning(ZSL)
stealth target