摘要
包络对齐是逆合成孔径雷达(Inverse Synthetic Aperture Radar,ISAR)成像中平动补偿处理的第一步,包络对齐的精度对于方位聚焦和成像质量具有重要影响。针对稀疏孔径和低信噪比条件下传统的包络对齐算法性能显著降低的问题,本文提出一种基于时空注意力-Seq2Seq网络的包络对齐方法。该网络模型以门控循环单元为编码解码单元,针对点目标距离像包络的能量分布特征对空间注意力机制进行改进后,添加时间和空间两维注意力机制形成对ISAR距离像回波包络进行对齐的能力。数据生成方面,基于电磁波仿真参数和目标运动仿真参数进行成像模拟仿真构造了ISAR回波数据集,经过8倍插值后输入网络进行训练,使网络学习到从未对齐回波到对齐回波的映射关系。所提方法以离线训练代替在线相关计算,融合了Seq2Seq模型在处理序列到序列问题上的结构优势、时间注意力机制在捕捉长期依赖关系和空间注意力机制在提取区域特征上的突出能力,实现了稀疏孔径和低信噪比条件下对距离-慢时间域ISAR回波的自动对齐。通过向训练好的时空注意力-Seq2Seq网络输入未对齐的回波序列,网络可以在不改变回波相位结构的前提下自动实现包络对齐。仿真和实测数据对齐结果表明,和传统的包络对齐方法相比,所提方法在稀疏孔径和低信噪比条件下优势明显,在欠采样率为50%、信噪比为0 dB条件下对雅克-42飞机实测回波数据的包络对齐实验中,该方法将循环移位误差由39、26减小至6,将成像结果的图像熵由4.58、4.22减小至1.71,验证了其良好性能。
Range alignment is the first step of translational compensation processing for inverse synthetic aperture radar(ISAR)imaging,and the accuracy of range alignment has a notable impact on the quality of azimuth focusing and final imaging.To solve the problem of the serious impairment of the performance of traditional range alignment algorithms un‐der the condition of sparse aperture and low signal-to-noise ratio(SNR),a novel range alignment method based on a spatiotemporal attention-sequence-to-sequence(Seq2Seq)network is proposed.A gated recurrent unit(GRU)was ad‐opted in this model as the encoding and decoding unit.The spatial attention mechanism was modified according to the unique energy distribution characteristics of the range profile of point-targets.The ability to align the ISAR range profile was finally formed by incorporating the temporal and spatial attention mechanism.For training data generation,an ISAR echo dataset was constructed through imaging simulation based on electromagnetic wave simulation parameters and target motion simulation parameters.After 8-fold interpolation,it was input into the network for training,allowing the network to learn the mapping relationship from unaligned echoes to aligned echoes.The proposed method replaced online correlation calculations with offline training.By integrating the advantages of the Seq2Seq network model in han‐dling Seq2Seq problems,the advantages of the temporal attention mechanism in capturing long-term dependencies,and the advantages of the spatial attention mechanism in extracting regional features,the proposed method achieved auto‐matic alignment of ISAR echoes in the range slow-time domain under sparse aperture and low SNR conditions.By input‐ting unaligned echo sequences into the trained spatiotemporal attention-Seq2Seq network,range alignment could be au‐tomatically achieved without changing the echo phase structure.Simulation and experimental data show that,compared with traditional range alignment methods,the proposed method obtained b
作者
李文哲
李开明
岳屹峰
王金昊
许慧革
罗迎
LI Wenzhe;LI Kaiming;YUE Yifeng;WANG Jinhao;XU Huige;LUO Ying(Unit 93107 of PLA,Shenyang,Liaoning 110141,China;College of Information and Navigation,Air Force Engineering University,Xi’an,Shaanxi 710077,China;Collaborative Innovation Center of Information Sensing and Understanding,Xi’an,Shaanxi 710077,China)
出处
《信号处理》
CSCD
北大核心
2024年第9期1659-1673,共15页
Journal of Signal Processing
基金
国家自然科学基金面上项目(62371468)
国家自然科学基金重点项目(62131020)。
关键词
逆合成孔径雷达成像
包络对齐
时空注意力机制
Seq2Seq模型
inverse synthetic aperture radar imaging
range alignment
spatiotemporal attention mechanism
sequence-to-sequence model
