| 引用本文: | 李宏博,吴文华,张云.基于ConvLSTM的空间进动锥体目标HRRP序列预测[J].哈尔滨工业大学学报,2023,55(10):10.DOI:10.11918/202207039 |
| LI Hongbo,WU Wenhua,ZHANG Yun.HRRP sequence prediction for spatial precession cone target based on ConvLSTM[J].Journal of Harbin Institute of Technology,2023,55(10):10.DOI:10.11918/202207039 |
|
| 摘要: |
| 宽带雷达对空间进动锥体目标进行持续探测能够形成高分辨距离像（high resolution range profile,HRRP）序列。HRRP序列携带空间进动锥体目标的空间几何信息和运动规律等信息,是进行目标关联跟踪和分类识别的重要依据,因此研究空间进动锥体目标的HRRP序列预测具有重要意义。卷积长短时记忆（ConvLSTM）网络将CNN和LSTM二者的特点有效结合,能够充分挖掘HRRP序列的空间和时间特性,完成对HRRP的预测。本文基于空间锥体目标的进动模型建立了多种尺寸、运动速度和运动方向等不同参数的HRRP序列数据集,并利用此数据集根据HRRP特性设计实现了适用于空间进动椎体目标HRRP预测的ConvLSTM网络模型。为了测试本文设计的ConvLSTM网络预测效果,将ConvLSTM网络与二维卷积神经网络模型进行预测效果对比分析。仿真实验结果表明,ConvLSTM网络预测结果与物理光学法计算得到的HRRP一致性高,皮尔逊相关系数高达0.973 1,平均绝对误差低至0.033 4,相较于二维卷积神经网络预测结果更加准确。证明本文设计的ConvLSTM网络模型能够有效提取HRRP序列的时间和空间特征,实现对HRRP序列的高精度预测。 |
| 关键词: 高分辨距离像 预测 ConvLSTM 空间目标 进动 |
| DOI：10.11918/202207039 |
| 分类号:TN975 |
| 文献标识码:A |
| 基金项目: |
|
| HRRP sequence prediction for spatial precession cone target based on ConvLSTM |
|
LI Hongbo1,2,WU Wenhua1,2,ZHANG Yun1,2
|
|
(1.School of Electronics and Information Engineering, Harbin Institute of Technology, Harbin, 150001, China; 2.Key Lab of Ocean Monitoring and Information Processing, Ministry of Industry and Information Technology(Harbin Institute of Technology), Harbin 150001, China)
|
| Abstract: |
| Continuous detection of space precession cone targets using broadband radars can generate high-resolution range profile (HRRP) sequences. The HRRP sequences contain information such as spatial geometric information and precession laws of space cone targets, which are applicable to target association, tracking and classification. Therefore, it is of great significance to perform HRRP sequence prediction for spatial precession cone targets. The ConvLSTM network effectively combines the characteristics of CNN and LSTM, which can fully mine spatial and temporal information of HRRP sequences to achieve prediction of HRRP. This article establishes a HRRP sequence dataset based on the precession model of spatial cone targets, which incorporates different parameters such as size, motion speed and motion direction, and uses this dataset to design and implement a ConvLSTM network model suitable for HRRP prediction for spatial precession cone targets according to HRRP characteristics. In order to validate the predictions by the ConvLSTM network designed in this paper, the ConvLSTM network is compared with the two-dimensional convolutional neural network mode. Simulation and experimental results show that ConvLSTM network is in good agreement with HRRP calculated using physical optics method, and is more accurate than predictions of 2D convolutionneural network. The Pearson correlation coefficient is as high as 0.973 1, and average absolute error reaches 0.033 4. The ConvLSTM network model can effectively extract temporal and spatial features of HRRP sequences to achieve high-precision prediction of HRRP sequences. |
| Key words: high resolution range profile prediction ConvLSTM spatial target precession |
