基于迁移学习的弹道目标ISAR图像识别方法  

Ballistic Target ISAR Image Recognition Method Based on Transfer Learning

作　　者：武文韬 WU Wen-tao(Air and Missile Defense College,Air Force Engineer University,Xianyang 718399,China)

机构地区：[1]空军工程大学防空反导学院,陕西咸阳713899

出　　处：《电脑与信息技术》2023年第2期27-30,共4页Computer and Information Technology

摘　　要：在反导作战预警阶段,需要尽快对敌方弹道导弹目标完成识别,ISAR图像是弹道目标识别的重要特征来源。传统ISAR成像识别方法包含特征提取和分类器设计等步骤,存在样本量少、识别准确率低等问题。鉴于此,提出了一种基于迁移学习的ISAR图像识别方法。首先基于转台成像模型生成假想目标在不同角度下得到的ISAR像,再利用三种预训练网络对已分组的数据进行训练,最后微调训练参数,获得最好的训练效果。实验结果表明,迁移学习能在样本量较少的情况下快速完成训练且具备较高的准确率,为弹道目标识别问题提供了新的解决方法。In the early warning stage of anti-missile operations,it is necessary to identify enemy ballistic missile targets as soon as possible,ISAR image is an important feature source for ballistic target recognition.The traditional ISAR imaging recognition method includes the steps of feature extraction and classifier design,but has the problems of small sample size and low recognition accuracy.In view of this,an ISAR image recognition method based on transfer learning is proposed.In view of this,an ISAR image recognition method based on transfer learning is proposed.Firstly,ISAR images obtained from different angles of the imaginary target are generated based on the turntable imaging model.Then,three kinds of pre-training networks are used to train the grouped data.Finally,the training parameters are fine-tuned to obtain the best training effect.The experimental results show that transfer learning can complete the training quickly and with high accuracy under the condition of small sample size,which provides a new solution for the problem of ballistic target recognition.

关 键 词：弹道导弹 目标识别 逆合成孔径雷达 深度神经网络 迁移学习 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]