航空学报 > 2023, Vol. 44 Issue (17): 328172-328172   doi: 10.7527/S1000-6893.2023.28172

结合迁移学习基于全卷积神经网络的ISAR自聚焦算法

王莲子, 汪玲(), 朱岱寅   

  1. 南京航空航天大学 电子信息工程院,南京 210016
  • 收稿日期:2022-10-26 修回日期:2022-11-17 接受日期:2023-01-11 出版日期:2023-09-15 发布日期:2023-02-10
  • 通讯作者: 汪玲 E-mail:tulip_wling@nuaa.edu.cn
  • 基金资助:
    国家自然科学基金(61871217);航空科学基金(20182052011);上海航天科技创新基金(SAST2021-026);南京航空航天大学科学前瞻布局科研专项(ILA220581A22)

An ISAR autofocus imaging algorithm based on FCN and transfer learing

Lianzi WANG, Ling WANG(), Daiyin ZHU   

  1. College of Electronic and information Engineering,Nanjing University of Aeronautics and Astronautics,Nanjing 210016,China
  • Received:2022-10-26 Revised:2022-11-17 Accepted:2023-01-11 Online:2023-09-15 Published:2023-02-10
  • Contact: Ling WANG E-mail:tulip_wling@nuaa.edu.cn
  • Supported by:
    National Natural Science Foundation of China(61871217);Aeronautical Science Foundation of China(20182052011);Shanghai Aerospace Science and Technology Innovation Fund(SAST2021-026);Fund of Prospective Layout of Scientific Research for NUAA (Nanjing University of Aeronautics and Astronautics)(ILA220581A22)

摘要:

逆合成孔径雷达(ISAR)成像目标多为非合作目标,在目标相对运动状态未知情况下的运动补偿是ISAR成像的关键。基于全卷积网络(FCN)具有强大的数据抽象特征挖掘和拟合能力,将FCN用于ISAR自聚焦,提出一种结合迁移学习基于FCN的ISAR自聚焦算法。通过构造大量不同相位误差的仿真数据集并进行训练,使FCN具有相位补偿能力,对不同姿态下仿真以及实测数据进行迁移训练,进一步提升平动相位补偿的能力。实测数据处理结果表明,结合迁移学习基于FCN的自聚焦成像效果优于传统类自聚集算法,验证了所提算法的有效性,相比传统方法更具优越性。

关键词: 逆合成孔径雷达, 成像, 相位补偿, 全卷积网络, 迁移学习

Abstract:

The targets of Inverse Synthetic Aperture Radar (ISAR) imaging are mostly non-cooperative, and motion compensation is the key of ISAR imaging when the relative motion status of target is unknown. Fully Convolutional Network (FCN) has excellent feature extraction and fitting abilities, and an ISAR phase compensation algorithm based on FCN and Transfer Learning is proposed by utilizing the FCN in ISAR autofocusing in this paper. Firstly, a large number of simulation data sets with different phase errors are constructed and trained to provide phase compensation ability for FCN. In addition, the transfer training of simulation and measured data with different attitudes is carried out to improve the translational phase compensation ability. The measured data processing results show that the performance of FCN phase compensation algorithm is better than traditional algorithm, which verifies the effectiveness in the proposed method and the superiority compared with the traditional method.

Key words: inverse synthetic aperture radar, imaging, phase compensation, fully convolutional neural network, transfer learning

中图分类号: