西安电子科技大学学报 ›› 2019, Vol. 46 ›› Issue (6): 125-130.doi: 10.19665/j.issn1001-2400.2019.06.018

• • 上一篇    下一篇

采用GRU模型的卫星RCS异常检测

胡盟霄,卢旺,徐灿,来嘉哲   

  1. 航天工程大学 宇航科学与技术系 北京 101416
  • 收稿日期:2019-07-06 出版日期:2019-12-20 发布日期:2019-12-21
  • 作者简介:胡盟霄(1995—),男,航天工程大学硕士研究生,E-mail:1256064775@qq.com
  • 基金资助:
    国家部委科技卓越青年科学基金(2017-JCJQ-005)

Satellite RCS anomaly detection using the GRU model

HU Mengxiao,LU Wang,XU Can,LAI Jiazhe   

  1. Dept of Aerospace Science and Technology, Space Engineering University,Beijing 101416,China
  • Received:2019-07-06 Online:2019-12-20 Published:2019-12-21

摘要:

针对传统基于雷达散射截面积的卫星目标姿态异常检测方法中提取有效特征困难、识别效果差的问题,提出了一种采用门控循环单元深度神经网络模型的异常检测方法。该方法首先利用滑动窗口法划分动态雷达散射截面积序列;然后采用门控循环单元深度神经网络完成对输入序列的自适应特征学习;最后结合全连接层实现卫星姿态异常检测。仿真实验结果表明,该方法提取的特征区分度高,与传统方法相比可以有效地检测出失稳翻滚卫星,并具有较强的噪声鲁棒性。

关键词: 门控循环单元网络, 雷达散射截面积, 卫星目标, 异常检测

Abstract:

Aiming at the problem that the traditional target image anomaly detection method based on radar cross-section extracts effective features with difficulty and the recognition effect is poor, an anomaly detection method gated recurrent unit deep neural network model is proposed. First, the method uses the sliding window method to divide the dynamic radar cross-sectional sequence. Then, to complete the adaptive feature learning of the input sequence, the gated recurrent unit deep neural network is used. Finally, the full connection layer is used to realize the satellite attitude anomaly detection. Simulation results show that the proposed method can achieve a high feature discrimination degree, that it can effectively detect the unstable rolling satellite compared with the traditional method, and that it has strong noise robustness.

Key words: gated recurrent unit, radar cross-sectional, satellite target, anomaly detection

中图分类号: 

  • TN957.52