-
微波关联成像借鉴光学关联成像思想,通过对发射信号调制,形成在时间和空间上具有起伏特性的随机辐射场,以模拟随机涨落特性的赝热光场分布,采用单阵元天线接收目标回波信号,将目标散射回波与随机辐射场进行关联处理,实现目标重构,成像原理如图2所示[15]。
不失一般性,考虑二维成像问题,设采用阵列方式对N个阵元发射信号幅度、频率或相位等参数进行调制,在空域合成具有强度随机涨落特性的辐射场分布,第
$n$ 个阵元发射信号记为$ {S_{{\text{T}}n}}(t) $ ,发射天线位置矢量记为${{{R}}_n}$ 。采用单阵元天线接收目标回波信号,接收信号记为$ {S_{\text{R}}}(t) $ ,接收天线位置矢量记为${{{R}}_{\boldsymbol{r}}}$ 。理想情况下,不同阵元发射信号相互正交,同一阵元发射信号在时间上不相关,即时间和空间相关函数满足[9, 16-17]:$$ R({t_1},{t_2}) = \int {{S_{{\text{T}}n}}(t - {t_1}){S_{{\text{T}}n}}(t - {t_2})} {\rm{d}}t = \delta ({t_1} - {t_2}) $$ (1) $$ R(n,m) = \int {{S_{{\text{T}}n}}(t){S_{{\text{T}}m}}(t)} {\rm{d}}t = \delta (n - m) $$ (2) 在成像区域
$ {\text{I}} $ 内,位置${{r}}$ 处的辐射信号记为${S_{\text{I}}}(t,{{r}})$ ,其中${{r}} \in {\text{I}}$ 。${S_{\text{I}}}(t,{{r}})$ 为${{N}}$ 个阵元发射信号在${{r}}$ 处的延时叠加,即:$$ {S_{\text{I}}}(t,{{r}}) = \sum\limits_{n = 1}^N {{S_{{\text{T}}n}}(t - \frac{{|{{r}} - {{{R}}_n}|}}{{c}})} $$ (3) 式中:c为光速。根据发射信号性质(1)和(2),辐射信号
${S_{\text{I}}}(t,{{r}})$ 关于时间和空间的二维相关函数为:$$ {R_{\text{I}}}(\tau ,\tau ';{{r}},{{r}}') = \int {{S_{\text{I}}}(t - \tau ,{{r}}){S_{\text{I}}}^ * (t - \tau ',{{r}}')} {\rm{d}}t \approx {{N}} \cdot \delta \left( {\tau - \tau ';{{r}} - {{r}}'} \right) $$ (4) 式中:“*”表示复信号的共轭。公式(4)表示在成像区域形成具有时间和空间独立性的辐射场分布。
目标回波信号可表示为:
$$ {S_{\text{R}}}(t) = \int_{\text{I}} {{\sigma _{{r}}}{S_{\text{I}}}\left(t - \frac{{|{{r}} - {{{R}}_r}|}}{{c}},{{r}}\right){\rm{d}}{{r}}} $$ (5) 式中:
$ {\sigma _r} $ 表示成像区域位置${{r}}$ 处的目标散射系数。将公式(5)中包含回波时延的辐射信号
$ {S_{\text{I}}} $ 记为关联成像参考信号,即:$$ S(t,{{r}}) = {S_{\text{I}}}\left(t - \frac{{|{{r}} - {{{R}}_r}|}}{{c}},{{r}}\right) $$ (6) 则目标回波信号可简化表示为:
$$ {S_{\text{R}}}(t) = \int_{\text{I}} {{\sigma _{{r}}}S(t,{{r}}){\rm{d}}{{r}}} $$ (7) 将目标回波信号与参考信号进行关联处理可得:
$$ \int {{S_{\text{R}}}(t){S^*}(t,{{r}})} {\rm{d}}t \approx {{N}} \cdot {\sigma _{{r}}} $$ (8) 因此,可以求得目标散射系数为:
$$ {\sigma _{{r}}} \approx \frac{1}{{{N}}}\int {{S_{\text{R}}}(t){S^*}(t,{{r}})} {\rm{d}}t $$ (9) 如公式(9)所示,在理想情况下,通过接收信号和参考信号之间的关联处理可以获得成像区域内任意位置处的散射信息。对于微波关联成像,由于各阵元发射信号已知,参考信号可根据发射信号以及成像几何计算得到,通过关联处理重构目标图像。与光学关联成像相比,微波关联成像可通过测量记录或计算方式获得参考信号,成像过程可通过计算方式实现。因此,微波关联成像可认为是一种计算关联成像方式。
若将时间和空间参数进行离散化表示,公式(7)也可以表示为离散形式。根据目标等效散射中心理论,光学区目标散射特性可等效为多个强散射点,将成像区域均匀划分为
${{L}}$ 个成像单元,每个单元网格中心位置矢量记为${{{r}}_l}$ ,$l = 1, \ldots ,{{L}}$ ,每个成像网格的目标等效散射系数记为${\sigma _l}$ 。将发射信号进行离散化,采样时刻记为${{t}} = {[{t_0},{t_1} \cdots {t_{{{M}} - 1}}]^{\rm T}}$ ,${{M}}$ 表示采样点数。则公式(7)可表示为矩阵形式:$$ \begin{split} &\left[ {\begin{array}{*{20}{c}} {{S_{\text{R}}}({t_0})} \\ {{S_{\text{R}}}({t_1})} \\ {...} \\ {{S_{\text{R}}}({t_{{{M}} - 1}})} \end{array}} \right] =\\ &\left[ {\begin{array}{*{20}{c}} {S({t_0},{{{r}}_1})}&{S({t_0},{{{r}}_2})}&{...}&{S({t_0},{{{r}}_{{L}}})} \\ {S({t_1},{{{r}}_1})}&{S({t_1},{{{r}}_2})}&{...}&{S({t_1},{{{r}}_{{L}}})} \\ {...}&{...}&{...}&{...} \\ {S({t_{{{M}} - 1}},{{{r}}_1})}&{S({t_{{{M}} - 1}},{{{r}}_2})}&{...}&{S({t_{{{M}} - 1}},{{{r}}_{{L}}})} \end{array}} \right] \cdot \left[ {\begin{array}{*{20}{c}} {{\sigma _1}} \\ {{\sigma _2}} \\ {...} \\ {{\sigma _{{L}}}} \end{array}} \right] \end{split}$$ (10) 若回波信号中存在测量噪声,则微波关联成像数学模型可表示为矢量形式:
$$ {{\boldsymbol{S}}_{\text{R}}} = {\boldsymbol{S}} \cdot {\boldsymbol{\sigma }} + {\boldsymbol{n}} $$ (11) 式中:
$ {{\boldsymbol{S}}_{\text{R}}} $ 表示回波信号矢量;$ {\boldsymbol{S}} $ 表示辐射场参考矩阵;$ {\boldsymbol{\sigma }} $ 表示目标散射系数矢量;$ {\boldsymbol{n}} $ 表示测量噪声矢量。回波信号矢量$ {{\boldsymbol{S}}_{\text{R}}} $ 由接收阵元采集数据得到,辐射场参考矩阵$ {\boldsymbol{S}} $ 由发射信号推演计算得出,目标散射系数矢量$ {\boldsymbol{\sigma }} $ 为待求解未知量。因此,微波关联成像可以建模为一个线性模型,成像重构可以视为线性逆问题的求解。根据辐射场参考矩阵
$ {{{\boldsymbol{S}}}} $ 以及目标散射系数矢量${{{\boldsymbol{\sigma}} }}$ 的性质,可以采用多种求解方法获得目标散射系数矢量${{{\boldsymbol{\sigma}}}}$ 的估计,从而重构目标图像。成像数学模型(11)通常也称为微波关联成像方程。 -
根据微波成像理论,传统微波成像的距离和方位向空间分辨率分别由发射信号带宽和天线有效孔径决定,即
$$ \left\{ \begin{gathered} {\rho _r} = \frac{{c}}{{2B}} \hfill \\ {\rho _\varphi } = \frac{\lambda }{D} \cdot R \hfill \\ \end{gathered} \right. $$ (12) 式中:B为发射信号带宽;
$ c $ 为光速;$ \lambda $ 为发射信号波长;D为成像的有效天线孔径;R为成像距离。当发射信号波长确定时,方位向分辨率取决于天线孔径尺寸。由于微波关联成像原理与传统微波成像技术有较大不同,成像分辨率与辐射场参考矩阵
$ {\boldsymbol{S}} $ 、目标散射系数分布$ {\boldsymbol{\sigma }} $ 以及求解方法等都存在联系。因此,难以给出解析的分辨率表达式,不过,通过对各种影响因素的分析,可以给出理论估计值。微波关联成像分辨率表征示意图如图3所示。其中,传统成像技术在空域发射相干信号,不同方位向信号具有较强相关性,而关联成像在空域发射时间和空间不相关的辐射信号,不同方位向信号相关性较弱。为了对成像分辨率进行定量分析,定义方位向两个相邻位置
$ {{{r}}_k} $ 和$ {{{r}}_{k'}} $ 的空间相关函数为[15]:图 3 微波关联成像分辨率表征[15]。(a) 传统成像相干发射情形;(b) 关联成像非相干发射情形;(c) 传统成像空间相关函数;(d) 关联成像空间相关函数
Figure 3. Resolution analysis of microwave coincidence imaging[15]. (a) Coherent transmissions of conventional imaging; (b) Incoherent transmissions of coincidence imaging; (c) Spatial correlation function of conventional imaging; (d) Spatial correlation function of coincidence imaging
$$ \chi \left( {{{{r}}_k},{{{r}}_{k'}}} \right) = \frac{{\left| {\left\langle {S\left( {t,{{{r}}_k}} \right),S\left( {t,{{{r}}_{k'}}} \right)} \right\rangle } \right|}}{{\left\| {S\left( {t,{{{r}}_k}} \right)} \right\|\left\| {S\left( {t,{{{r}}_{k'}}} \right)} \right\|}} $$ (13) 式中:
$\left\langle {S\left( {t,{{{r}}_k}} \right),S\left( {t,{{{r}}_{k'}}} \right)} \right\rangle$ 表示关于时间$ t $ 的集合平均;$ \left\| \cdot \right\| $ 表示矩阵范数。空间相关函数给出了辐射场参考信号
$S(t,{{r}})$ 时间和空间相关性的度量,与成像分辨率存在紧密联系。当两个相邻位置的参考信号$ S\left( {t,{{{r}}_k}} \right) $ 和$ S\left( {t,{{{r}}_{k'}}} \right) $ 完全相关时,空间相关函数$ \chi \left( {{{{r}}_k},{{{r}}_{k'}}} \right) $ 取最大值,当二者完全不相关时,空间相关函数$ \chi \left( {{{{r}}_k},{{{r}}_{k'}}} \right) = 0 $ 。在理想情况下,当辐射场完全随机起伏时,空间相关函数为冲激函数$ \chi \left( {{{{r}}_k},{{{r}}_{k'}}} \right) = \delta \left( {{{{r}}_k} - {{{r}}_{k'}}} \right) $ 。通常,辐射场空间相关函数随方位向距离$ d = \left\| {{{{r}}_k} - {{{r}}_{k'}}} \right\| $ 增大而减小,其主瓣宽度与关联成像空间分辨率有关。为了描述关联成像空间分辨率,辐射场空间相关函数的主瓣宽度可由门限
${\chi _{{\rm{th}}}}$ 确定。若$\chi \left( {{{{r}}_k},{{{r}}_{k'}}} \right) > {\chi _{{\rm{th}}}}$ ,则两个相邻散射点信号具有较强相关性,不能分辨。临界分辨率可表示为:$$ {\rho _\varphi } = \left\| {{{{r}}_k} - {{{r}}_{k'}}} \right\| , {\text{s}}{\text{.t}}{\text{. }}\chi \left( {{{{r}}_k},{{{r}}_{k'}}} \right) = {\chi _{{\rm{th}}}} $$ (14) 其中,成像固有分辨能力由发射信号形成的辐射场空间相关函数决定,而分辨率门限
${\chi _{{\rm{th}}}}$ 与目标散射系数分布以及成像方法有关。不同发射波形对辐射场起伏程度以及空间相关函数的影响不同。图4(a)~(f)给出了随机调频、随机调幅和随机调相三种典型波形形成的辐射场及空间相关函数图。其中,发射阵列为均匀线阵,阵元个数为10,阵元间距为0.5 m,发射信号中心频率为9.5 GHz,带宽为500 MHz,辐射场平面与发射阵列距离为1 km。从图中可以看出,随机调频和随机调幅波形形成的辐射场比随机调相波形具有更强的起伏特性,空间相关函数主瓣较窄,其成像分辨率较高。
图 4 不同发射波形的空间相关函数[9]。(a)~(b) 随机调频波形辐射场和空间相关函数;(c)~(d) 随机调幅波形辐射场和空间相关函数;(e)~(f) 随机调相波形辐射场和空间相关函数
Figure 4. Spatial correlation functions of transmitted waveforms[9]. (a)-(b) Radiation field and spatial correlation function of random frequency modulation waveform; (c)-(d) Radiation field and spatial correlation function of random amplitude modulation waveform; (e)-(f) Radiation field and spatial correlation function of random phase modulation waveform
除了采用空间相关函数表征微波关联成像的固有名义分辨率,还可以采用统计分辨力[17]、平均模糊函数[18]等定义微波关联成像分辨能力。基于目标位置估计,子空间投影分辨率表征方法[19]给出了成像分辨率的一般性表述,并对成像分辨率的各种影响因素进行分析。根据线性方程组求解理论,从对公式(11)表示的微波关联成像数学模型求解的角度出发,有效秩理论可以在一定程度上反映求解的性能及成像重构的超分辨性能。
-
理论研究表明,采用二阶关联处理比一阶关联处理成像分辨率提高约55%[19]。为了进一步提高微波关联成像分辨率,基于公式(11)的成像模型,已提出多种成像方法,通过减小成像单元尺寸,细分目标散射分布,获得超分辨性能。然而,当成像单元尺寸减小时,相邻网格辐射场之间的差异性减弱,相关性增大,矩阵方程趋于病态,将求解稳定性变差。辐射场参考矩阵的有效秩可反映辐射场的相关特性[17],定量表征信号参数、成像单元尺寸以及目标复杂度等对成像重构的影响及限制性要求。
图5给出了辐射场参考矩阵有效秩随成像单元尺寸变化的关系曲线。成像区域划分为
$ 30 \times 30 $ 个成像单元,成像单元尺寸$0.2\;{\text{m}} \leqslant \Delta \leqslant 2\;{\text{m}}$ 。增大成像单元尺寸,相邻单元辐射场信号差异性提高,辐射场参考矩阵有效秩增大,当有效秩与成像单元个数相近时,辐射场参考矩阵为满秩,此时独立方程个数等于目标散射系数参数个数,目标图像可通过矩阵求逆稳健重构,此时成像分辨率由成像单元尺寸确定。理论上,提高工作频率、增大天线孔径、增大发射信号带宽,均可提高辐射场的空间差异性,增大辐射场参考矩阵有效秩,从而获得更高成像分辨率。 -
微波关联成像数学模型表明,成像重构是一个由矩阵方程描述的线性逆问题求解过程。由于实际的成像系统难以产生具有完全不相干特性的辐射场,而观测向量维度M通常小于成像单元个数L,因此,微波关联成像模型通常是一个病态的逆问题[20]。在不存在模型误差的理想情况下,理论上可以采用相关法、最小二乘法、正则化方法和压缩感知等基本成像方法重构目标图像。在实际情况下,当存在系统误差、信号误差、成像单元网格失配、目标运动失配等模型失配误差时,需要克服失配影响,存在各种模型失配情况下的稳健重构方法是微波关联成像算法研究的一个重点。
-
根据关联成像思想,目标散射系数矢量可以通过关联处理得到,即
$$ \hat{{\boldsymbol\sigma} }={{\boldsymbol{S}}}^{{\rm{H}}}\cdot{{\boldsymbol{S}}}_{\text{R}} $$ (15) 式中:
${\rm{H}}$ 表示矩阵的共轭转置。公式(15)所示的相关法直接将接收信号与辐射场参考矩阵进行相关,属于一阶关联处理,与传统微波成像中的匹配滤波处理一致。相关法可以避免由成像方程的病态所导致的求解不稳定现象,理论上成像结果无法获得超分辨效果。当存在测量噪声时,${{\boldsymbol{S}}}^{{\rm{H}}}\cdot{{\boldsymbol{S}}}_{\text{R}}\approx {{\boldsymbol{S}}}^{{\rm{H}}}\cdot\left({{\boldsymbol{S}}}_{\text{R}}+{\boldsymbol{n}}\right)$ ,由于测量噪声通常与辐射场不相关,所以相关法的优势是对测量噪声的容忍能力较强,成像结果较稳健。 -
最小二乘估计将成像模型转化为一个最小二乘问题进行求解,即:
$$ \hat{{{\boldsymbol\sigma}} }=\arg\underset{{{\boldsymbol\sigma}} }{\min}{\Vert {{{\boldsymbol S}}}_{\text{R}}-{{\boldsymbol S}}\cdot{{\boldsymbol\sigma}} \Vert }_{2}^{2} $$ (16) 最小二乘法得到的目标散射系数估计量可表示为[21]:
$$ \hat{{{\boldsymbol\sigma}} }={{{\boldsymbol S}}}^{\dagger}\cdot{{{\boldsymbol S}}}_{\text{R}} $$ (17) 式中:
${{{\boldsymbol S}}^\dagger } = {\left( {{{{\boldsymbol S}}^{\rm{H}}}{{\boldsymbol S}}} \right)^{ - 1}}{{{\boldsymbol S}}^{\rm{H}}}$ 是${{\boldsymbol S}}$ 的伪逆[22]。最小二乘法可以提高目标的成像分辨率,但对模型误差较敏感。此外,最小二乘法的性能也受到辐射场特性的约束。对于线性模型(16),在高斯噪声情况下,最小二乘估计(17)为最佳估计。最小二乘法具有一定的抗噪能力,但是要利用公式(17)对方程求解的一个必要条件是矩阵
${{\boldsymbol{S}}^{\rm{H}}}{\boldsymbol{S}}$ 可逆,当辐射场参考矩阵${\boldsymbol{S}}$ 不是满秩、条件数(Condition number)较大时,成像效果不理想。辐射场参考矩阵${\boldsymbol{S}} $ 的条件数$ {N_{{\text{cond}}}} $ 定义为[23]:$$ {N}_{\text{cond}}=\Vert {{\boldsymbol S}}\Vert \cdot\Vert {{{\boldsymbol S}}}^{-1}\Vert =\frac{{\xi }_{\mathrm{max}}}{{\xi }_{\mathrm{min}}} $$ (18) 式中:
$ {\xi _{\max }} $ 和$ {\xi _{\min }} $ 分别表示${{\boldsymbol S}}$ 的最大和最小奇异值。辐射场参考矩阵
${{\boldsymbol S}}$ 的条件数取决于辐射场的非相干特性,条件数越大,非相干性越弱。空间上距离越靠近的两个位置,辐射场的相关性越强,因此当成像单元划分越细时,对应的矩阵条件数越大,导致最小二乘算法求解不稳定。条件数大的参考矩阵对应的成像方程大多属于病态方程,此时测量数据的微小扰动都可能引起求解结果产生较大误差。而条件数接近1时矩阵是良态的,此时方程求解误差对模型扰动敏感性较弱。因此,最小二乘法适用于辐射场参考矩阵${\boldsymbol{S}}$ 非相干特性较好的情况,而相关法则是对最小二乘方法中的不稳定部分作了近似处理,使求解更加稳定[24]。 -
在实际成像处理中,辐射场参考矩阵
${{\boldsymbol S}}$ 通常不是满秩的,最小二乘法的求解效果不理想,主要原因是矩阵${{{\boldsymbol S}}^{\rm{H}}}{{\boldsymbol S}}$ 求逆不稳定。作为对最小二乘法代价函数${\Vert {{{\boldsymbol S}}}_{\text{R}}-{{\boldsymbol S}}\cdot{{\boldsymbol\sigma}} \Vert }_{2}^{2}$ 的改进,Tikhonov正则化方法修正了代价函数,$$ J({{\boldsymbol\sigma}} )=\mathrm{arg}\;\underset{{{\boldsymbol\sigma}} }{\mathrm{min}}\left({\Vert {{{\boldsymbol S}}}_{\text{R}}-{{\boldsymbol S}}\cdot{{\boldsymbol\sigma}} \Vert }_{2}^{2}+\eta {\Vert {{\boldsymbol\sigma}} \Vert }_{2}^{2}\right) $$ (19) 式中:
$ \eta $ 称为正则化参数,求解方法称为Tikhonov正则化法[25-26]。其解可表示为:$$ {\hat{\boldsymbol \sigma }} = {\left( {{{{\boldsymbol S}}^{\rm{H}}}{{\boldsymbol S}} + \eta {{I}}} \right)^{ - 1}}{{{\boldsymbol S}}^{\rm{H}}}{{\boldsymbol S}_{\text{R}}} $$ (20) Tikhonov正则化方法的本质是通过对秩亏矩阵
${{{\boldsymbol S}}^{\rm{H}}}{{\boldsymbol S}}$ 每一个对角元素加一个微小的扰动参数$ \eta $ ,使得对接近奇异的协方差矩阵${{{\boldsymbol S}}^{\rm{H}}}{{\boldsymbol S}}$ 求逆变成非奇异矩阵${{{\boldsymbol S}}^{\rm{H}}}{{\boldsymbol S}} + \eta {{I}}$ 的求逆,从而提高求解的稳定性。 -
电磁散射理论表明,目标在高频区的后向散射特性可由散射中心模型描述[27],目标的散射特性可以近似为某些局部位置的散射中心响应的合成,这些局部性的散射源称为目标等效散射中心。目标等效散射中心通常具有稀疏分布特性,利用目标散射的稀疏性可以提高成像质量。在数学上,微波关联成像模型与压缩感知模型具有相似性,因此,采用压缩感知方法实现目标图像重构也是关联成像处理的一个途径。这类算法利用目标散射中心分布的稀疏先验,在测量数据有限的情况下,准确求解出
${{\boldsymbol\sigma }}$ ,此外还可以克服相关法宽主瓣和高旁瓣问题,抑制虚假散射点,提高成像质量。根据压缩感知理论,若$ {\boldsymbol{\sigma }} $ 稀疏,以下优化问题能准确求解[28]:$$ \hat{{{\boldsymbol\sigma}} }=\mathrm{arg}\;\underset{{{\boldsymbol\sigma}} }{\mathrm{min}}\left\{{\Vert {{{\boldsymbol S}}}_{\text{R}}-{{\boldsymbol S}}\cdot{{\boldsymbol\sigma}} \Vert }_{2}^{2}+\eta {\Vert {{\boldsymbol\sigma}} \Vert }_{0}\right\} $$ (21) 然而,上述
$ {\ell _0} $ 范数求解是一个NP难问题[29]。通过近似处理能够求解该问题,最常用的方法是用$ {\ell _1} $ 范数代替$ {\ell _0} $ 范数[28, 30]:$$ \hat{{{\boldsymbol\sigma}} }=\mathrm{arg}\;\underset{{{\boldsymbol\sigma}} }{\mathrm{min}}\left\{{\Vert {{{\boldsymbol S}}}_{\text{R}}-{{\boldsymbol S}}\cdot{{\boldsymbol\sigma}} \Vert }_{2}^{2}+\eta {\Vert {{\boldsymbol\sigma}} \Vert }_{1}\right\} $$ (22) 公式(22)是一个凸优化问题[31],其思想是在满足成像方程约束的前提下,求得目标稀疏解。当存在测量噪声时,问题转化为成像误差在一定范围内,使目标最稀疏。压缩感知重构方法通常具有较高的计算复杂度。
当存在测量噪声时,采用贝叶斯方法通常能够降低噪声影响,在逆问题求解时以概率形式考虑因测量噪声带来的不确定性和误差,利用先验信息和测量数据获得对未知参数的估计。贝叶斯方法与稀疏表示理论结合而发展的稀疏贝叶斯方法能够充分利用测量数据的先验信息,在目标散射系数服从某一稀疏先验分布
$p\left( {\boldsymbol{\sigma }} \right)$ 时,根据贝叶斯原理得出散射系数的后验概率估计。稀疏贝叶斯重构方法的目标函数可以表示为:$$ \hat{{{\boldsymbol\sigma}} }=\mathrm{arg}\;\underset{{{\boldsymbol\sigma}} }{\mathrm{max}}\;\mathrm{log}\left(p\left({{{\boldsymbol S}}}_{\text{R}}|{{\boldsymbol\sigma}} \right)\cdot p\left({{\boldsymbol\sigma}} \right)\right) $$ (23) 微波关联成像中常用的压缩感知求解方法主要有正交匹配追踪(Orthogonal matching pursuit, OMP)算法[32]、基追踪(Basis persuit, BP)算法[31]、稀疏贝叶斯学习(Sparse Bayesian learning, SBL)算法[33]等。图6给出了三种不同目标复杂度情况下,相关法、最小二乘法、Tikhonov正则化方法和SBL成像结果。其中,相关法和最小二乘法的成像质量较差,Tikhonov正则化法对矩阵求逆过程做了改进,成像质量有所提高。当目标稀疏度较高时,SBL法的成像结果较好,但不适用于复杂扩展目标成像。
-
第3.1节中所述基本成像算法只考虑了测量噪声影响,适用于关联成像模型准确情况下的目标重构。在实际中,系统误差、目标运动等会造成成像模型误差,模型失配对成像重构的准确性和成像质量有较大影响,是关联成像处理中的一个重要问题。
从微波关联成像处理过程可知,在关联处理前需要首先获取辐射场参考矩阵,这就需要成像模型各参数精确已知。但在实际成像过程中,普遍存在各种模型误差,如阵元间时间同步误差[34-35]、阵元位置误差[36-37]、信号幅相误差[38-42]、目标散射中心网格失配误差[43-44]和目标运动引起的失配误差[18]等,如图7所示。相比于加性噪声,模型误差引起的辐射场参考矩阵误差是一类乘性误差,建模更加复杂,对成像质量的影响也更大。
(1)针对阵元间时间同步误差,通过分析随机辐射场参考信号和回波之间的时间同步误差对微波关联成像质量的影响,基于随机辐射场短时积累的微波关联成像方法采用“脉内积累、关联处理”方式降低了时间同步误差带来的不利影响,提高了成像质量[34]。基于分时累积的图像重构方法则通过一定时间的辐射场空间差异性积累,降低了对时间同步精度的要求[35]。
(2)针对阵元位置误差,通过建立存在阵元位置误差的微波关联成像模型,推导存在阵元位置误差的散射系数估计下界CRLB,同时结合参数估计和辅助物校正、自校正等方法,在稀疏重构框架下的辅助目标校正法、迭代补偿以及等效补偿等方法能够较好地解决阵元位置误差带来的影响[16, 36-37]。
(3)针对信号幅相误差,基于凸优化理论和相位恢复算法,基于压缩感知的阵列幅相误差估计方法可对信号幅度误差和相位误差进行估计和校正[38]。基于辅助阵元的微波关联成像幅相误差校正方法,能够分别估计幅度和相位误差并进行补偿,从而获得准确的辐射场参考矩阵[42]。
(4)针对目标散射中心网格失配误差,已提出多种成像方法。在关联成像处理时,通常将成像平面划分为若干个成像单元,并假设目标散射中心位于成像单元网格中心处,但实际目标散射中心不可避免与划分的虚拟网格中心存在偏差[43-44],从而形成成像模型误差。针对网格失配问题,相关法与参数化方法联合重构能够降低模型失配误差的敏感性[9, 44]。从贪婪迭代求解和贝叶斯统计优化思路出发,基于迭代
$ {\ell _0} $ 范数的最小二乘成像算法和基于迭代最大后验的稀疏自适应校正反演方法能够对网格失配误差进行有效估计和校正[8]。前者是采用多分辨率策略更新辐射场矩阵,达到搜索校正网格失配误差的效果,后者是通过赋予散射系数和网格失配误差先验分布,从最大后验概率出发同时估计散射系数和失配误差。基于结构特征的稀疏总体最小二乘方法能够对网格失配进行校正,并利用FOCUSS方法对目标进行重构[45]。此外,不同于固定网格成像方法,多重网格划分成像方法[46-47]和动态网格划分成像方法[48-49]按照一定策略使网格迭代更新进化,从而能够使网格更好地聚集在散射点附近,有效减小网格失配误差并提高成像效率。基于瀑布型多重网格的处理方法首先对成像方程进行预处理,将原问题分解到多级疏密程度不同的网格空间,然后依次进行求解,以此来解决大场景下成像方程规模大的问题[46]。基于目标分布的非均匀网格处理方法对成像区域进行非均匀网格划分从而利用较少的网格实现对大成像场景的成像[47]。定向网格分裂成像方法将参考矩阵关于网格坐标的一阶导数加入成像模型中,并根据一阶导数确定新网格的位置,从而使网格位置更加靠近目标散射点,减小成像网格失配误差[48]。迭代重加权动态网格成像方法能够在迭代中不断细化网格,并将散射系数值作为权重值作用到网格上,从而能够剔除冗余网格,使网格聚集于目标散射点附近,在减小网格失配误差的同时提高成像效率[49],如图8所示。
(5)针对目标运动引起的失配误差,中国科学技术大学[8, 45, 50]和国防科技大学[9, 18]开展了相关算法研究。处理思路包括两种:一是基于高精度运动补偿的运动目标关联成像,通过对目标运动参数进行高精度估计,更新辐射场参考矩阵,实现运动目标重构;二是基于多维联合重构的运动目标关联成像,建立包含目标运动参数和目标散射系数的关联成像方程,通过迭代求解,对目标运动参数和目标散射系数进行联合估计。基于更新过完备字典的方法和基于速度估计的自适应稀疏反演方法能够较好地对运动目标进行成像[8];基于速度预估计的运动目标成像方法先采用相关法估计目标运动速度,然后对辐射场矩阵进行运动速度补偿[50];此外,基于参数化字典学习的自适应稀疏反演算法把对目标运动速度和目标位置的求解问题转换为连续参数估计问题,对经典SBL算法进行修改,引入运动目标速度的估计,迭代得到目标图像和运动速度[50-52]。对于静止平台,通过对目标运动状态分析获取目标速度和加速度信息,然后选择合适的观测矩阵进行成像[45];对于运动平台,关联成像体制下的目标角度获取和跟踪算法能够以先跟踪后成像的方式对动目标进行成像[18]。
Progress and prospect of microwave coincidence imaging(Invited)
-
摘要: 微波关联成像起源于光学强度关联成像,通过对电磁波的调控形成空变和时变的辐射模式,突破天线孔径对成像分辨率的限制,具有前视、凝视、快拍成像等优势,在重点区域凝视观测、无人系统自主感知、安检安防等领域具有广阔的应用前景。文中简述了微波关联成像的技术起源,从成像原理、成像方法、成像系统等三个方面,总结了微波关联成像的研究现状与主要进展。通过对成像原理的剖析,阐明关联成像的基本条件与成像分辨率的影响因素;通过对成像方法的梳理,分析微波关联成像与光学关联成像以及传统微波成像方法之间的区别与联系;通过对成像系统的介绍,比较随机辐射、波前调制、孔径编码等多种成像体制的特点与差异,厘清技术发展脉络。最后,总结并展望了微波关联成像的未来发展趋势。Abstract: Originated from the optical intensity ghost imaging, microwave coincidence imaging breaks through the limitation of antenna aperture on imaging resolution by space and time-varying radiation mode through the modulation of electromagnetic waves. It has the advantages of forward-looking, staring and fast-shooting imaging, and has broad application prospects in the fields of staring observation in key areas, autonomous sensing of unmanned systems, security inspection and security protection and so on. The technical origin of microwave coincidence imaging was briefly described, and its current research status and main progress from three aspects including the imaging principle, imaging methods and imaging systems were summarized. Through the analysis of the imaging principle, the basic conditions for coincidence imaging and the influencing factors of imaging resolution were clarified. Through the review of imaging methods, the differences and relationships among microwave coincidence imaging, optical ghost imaging and conventional microwave imaging methods were analyzed. Through the introduction of imaging systems, the features and differences among various systems such as random radiation, wavefront modulation and aperture encoding were compared, which made the development of microwave coincidence imaging easier to perceive. Finally, the future development trend of microwave coincidence imaging was summarized and prospected.
-
图 3 微波关联成像分辨率表征[15]。(a) 传统成像相干发射情形;(b) 关联成像非相干发射情形;(c) 传统成像空间相关函数;(d) 关联成像空间相关函数
Figure 3. Resolution analysis of microwave coincidence imaging[15]. (a) Coherent transmissions of conventional imaging; (b) Incoherent transmissions of coincidence imaging; (c) Spatial correlation function of conventional imaging; (d) Spatial correlation function of coincidence imaging
图 4 不同发射波形的空间相关函数[9]。(a)~(b) 随机调频波形辐射场和空间相关函数;(c)~(d) 随机调幅波形辐射场和空间相关函数;(e)~(f) 随机调相波形辐射场和空间相关函数
Figure 4. Spatial correlation functions of transmitted waveforms[9]. (a)-(b) Radiation field and spatial correlation function of random frequency modulation waveform; (c)-(d) Radiation field and spatial correlation function of random amplitude modulation waveform; (e)-(f) Radiation field and spatial correlation function of random phase modulation waveform
-
[1] Ausherman D A, Kozma A, Walker J L, et al. Developments in radar imaging [J]. IEEE Transactions on Aerospace and Electronic Systems, 1984, 20(4): 363-400. [2] Nan Y J, Huang X J, Guo Y J. Generalized continuous wave synthetic aperture radar for high resolution and wide swath remote sensing [J]. IEEE Transactions on Geoscience and Remote Sensing, 2018, 56(12): 7217-7229. doi: 10.1109/TGRS.2018.2849382 [3] Zhang L, Qiao Z J, Xing M D, et al. High-resolution ISAR imaging by exploiting sparse apertures [J]. IEEE Transactions on Antennas and Propagation, 2012, 60(2): 997-1008. doi: 10.1109/TAP.2011.2173130 [4] Yang J Y. Multi-directional evolution trend and law analysis of radar ground imaging technology [J]. Journal of Radars, 2019, 8(6): 669-693. (in Chinese) [5] Liu W T, Sun S, Hu H K, et al. Progress and prospect for ghost imaging of moving objects [J]. Laser and Optoelectronics Progress, 2021, 58(10): 3-16. (in Chinese) [6] Guo Y Y, Wang D J, He X Z, et al. Super-resolution imaging method based on random radiation radar array[C]//2012 IEEE International Conference on Imaging Systems and Techniques Proceedings, Manchester, 2012: 1-6. [7] Guo Y Y, He X Z, Wang D J. A novel super-resolution imaging method based on stochastic radiation radar array [J]. Measurement Science and Technology, 2013, 24(7): 074013. doi: 10.1088/0957-0233/24/7/074013 [8] He X Z. The information processing methods and simulations in microwave staring correlated imaging[D]. Hefei: University of Science and Technology of China, 2013. (in Chinese) [9] Li D Z. Radar coincidence imaging technique research[D]. Changsha: National University of Defense Technology, 2014. (in Chinese) [10] Shao Z L. Design on spatial-temporal random radiation field for compressed sensing based microwave imaging radar[D]. Xi’an: Xidian University, 2014. (in Chinese) [11] Xu R. Study on new systems and techniques for improving radar imaging performances[D]. Xi’an: Xidian University, 2015. (in Chinese) [12] Chen J P, Zhu W G, Zhang G. A new method of microwave relating imaging [J]. Journal of Naval Aeronautical and Astronautical University, 2012, 27(2): 196-198. (in Chinese) doi: 10.3969/j.issn.1673-1522.2012.02.016 [13] Shao P, Xu R, Li H L, et al. The research on bjorck-schmidt orthogonalization for microwave staring imaging [J]. Journal of Signal Processing, 2014, 30(4): 450-456. (in Chinese) [14] Zhu S T, Zhang A X, Xu Z, et al. Radar coincidence imaging with random microwave source [J]. IEEE Antennas Wireless Propagation Letters, 2015, 14: 1239-1242. doi: 10.1109/LAWP.2015.2399977 [15] Cheng Y Q, Zhou X L, Xu X W, et al. Radar coincidence imaging with stochastic frequency modulated array [J]. IEEE Journal of Selected Topics in Signal Processing, 2016, 8(5): 513-524. [16] Xu X W. Research on radar coincidence imaging with array position error[D]. Changsha: National University of Defense Technology, 2015. (in Chinese) [17] Zhou X L. Theory and methods of sparsity-based microwave coincidence imaging[D]. Changsha: National University of Defense Technology, 2017. (in Chinese) [18] Zha G F. Microwave coincidence imaging technique research for moving target[D]. Changsha: National University of Defense Technology, 2016. (in Chinese) [19] Zhu S T, He Y C, Chen X M, et al. Resolution threshold analysis of the microwave radar coincidence imaging [J]. IEEE Transactions on Geoscience and Remote Sensing, 2020, 58(3): 2232-2243. doi: 10.1109/TGRS.2019.2955789 [20] Wang T Y. Research on distributed radar sparse imaging technologies[D]. Hefei: University of Science and Technology of China, 2015. (in Chinese) [21] Kay S M. Fundamentals of Statistical Signal Processing: Estimation Theory[M]. Englewood: Prentice Hall, 1993. [22] Albert A. Regress and the Moore-Penrose Pseudoinverse[M]. New York: Academic Press, 1972. [23] Golub G H, Van Loan C F. Matrix Computations[M]. 3rd ed. Baltimore: Johns Hopkins University Press, 1996. [24] Yang J G. Research on sparsity-driven regularization radar imaging theory and method[D]. Changsha: National University of Defense Technology, 2014. (in Chinese) [25] Phillips D L. A technique for the numerical solution of certain integral equations of the first kind [J]. Journal of the Association for Computing Machinery, 1962, 9: 84-97. doi: 10.1145/321105.321114 [26] Tikhonov A N. Solution of incorrectly formulated problems and the regularization method [J]. Soviet Mathematics-Doklady, 1963, 4: 1035-1038. [27] Potter L C, Chiang D, Carriere R, et al. A GTD-based parametric model for radar scattering [J]. IEEE Transactions on Antennas Propagation, 1995, 32(10): 1058-1067. [28] Baraniuk R G. Compressive sensing [J]. IEEE Signal Processing Magazine, 2007, 24(4): 118-121. doi: 10.1109/MSP.2007.4286571 [29] Donoho D L. For most large underdetermined systems of linear equations, the minimal L1 norm solution is also the sparsest solution [J]. Communications on Pure and Applied Mathematics, 2006, 59(6): 797-829. doi: 10.1002/cpa.20132 [30] Candès E J, Wakin M B. An introduction to compressive sampling: a sensing/sampling paradigm that goes against the common knowledge in data acquisition [J]. IEEE Signal Processing Magazine, 2008, 25(2): 21-30. doi: 10.1109/MSP.2007.914731 [31] Chen S, Donoho D, Saunders M. Atomic decomposition by basis pursuit [J]. SIAM Review, 2001, 43(1): 129-159. doi: 10.1137/S003614450037906X [32] Tropp J A, Gilbert A C. Signal recovery from random measurements via orthogonal matching pursuit [J]. IEEE Transactions on Information Theory, 2007, 53(12): 4655-4666. doi: 10.1109/TIT.2007.909108 [33] Wipf D P, Rao B D. Sparse Bayesian learning for basis selection [J]. IEEE Transactions on Signal Processing, 2004, 52(8): 2153-2164. doi: 10.1109/TSP.2004.831016 [34] Guo Y, Ma Y, Wang D. A novel microwave staring imaging method based on short-time integral stochastic radiation fields[C]//2013 IEEE International Conference on Imaging Systems and Techniques, 2013: 425-430. [35] Ma Y P. Preliminary research on microwave staring correlated imaging based on temporal-spatial stochastic radiation fields[D]. Hefei: University of Science and Technology of China, 2013. (in Chinese) [36] Xu X W, Cheng Y Q, Qin Y L, et al. Analysis of array position error in radar coincidence imaging [J]. Modern Radar, 2016, 38(3): 32-37. (in Chinese) [37] Xu X W, Zhou X L, Cheng Y Q, et al. Radar coincidence imaging with array position error[C]//2015 IEEE International Conference on Signal Processing, Communications and Computing (ICSPCC 2015), 2015: 119-122. [38] Yi M L. Study on compressed sensing algorithm for microwave staring imaging[D]. Xi’an: Xidian University, 2014. (in Chinese) [39] Zhou X, Wang H, Cheng Y, et al. Sparse auto-calibration for radar coincidence imaging with gain-phase error [J]. Sensors, 2015, 15: 27611-27624. doi: 10.3390/s151127611 [40] Zhou X, Wang H, Cheng Y, et al. Radar coincidence imaging with phase error using Bayesian hierarchical prior modeling [J]. Journal of Electronic Imaging, 2016, 25(1): 013018. doi: 10.1117/1.JEI.25.1.013018 [41] Zhou X, Wang H, Cheng Y, et al. An ExCoV-based method for joint radar coincidence imaging and gain-phase error calibration [J]. Mathematical Problems in Engineering, 2016, 8(5): 513-524. [42] Zheng Y. Array self-calibration for MIMO radar with gain-phase error[D]. Xi’an: Xidian University, 2015. (in Chinese) [43] Zhou X, Wang H, Cheng Y, et al. Off-grid radar coincidence imaging based on block sparse Bayesian learning[C]//2015 IEEE Workshop on Signal Processing Systems (SiPS), 2015: 440-443. [44] Li D, Li X, Cheng Y, et al. Radar coincidence imaging under grid mismatch [J]. ISRN Signal Processing, 2014, 987803: 1-8. [45] Luo C S. Research on microwave correlated sparse imaging of moving target[D]. Hefei: University of Science and Technology of China, 2016. (in Chinese) [46] Wang G C. Research on microwave staring correlated imaging of low-rank and large scene[D]. Hefei: University of Science and Technology of China, 2018. (in Chinese) [47] Meng Q Q. The research on information processing in high resolution microwave staring correlated imaging[D]. Hefei: University of Science and Technology of China, 2016. (in Chinese) [48] Cao K C, Cheng Y Q, Liu K, et al. Off-grid microwave coincidence imaging based on directional grid fission [J]. IEEE Antennas Wireless Propagation Letters, 2020, 19(12): 2497-2501. doi: 10.1109/LAWP.2020.3037100 [49] Cao K C, Cheng Y Q, Liu K, et al. Reweighted-dynamic-grid-based microwave coincidence imaging with grid mismatch[J/OL]. IEEE Transactions on Geoscience and Remote Sensing(2021-06-15)https://ieeexplore.ieee.org/document/9455128/authors#authors. [50] Zhang H L. Research on sparse reconstruction technology for microwave staring correlated imaging of moving target[D]. Hefei: University of Science and Technology of China, 2015. (in Chinese) [51] Li D Z, Li X, Cheng Y Q, et al. Radar coincidence imaging in the presence of target-motion-induced error [J]. Journal of Electronic Imaging, 2014, 23(2): 023014. doi: 10.1117/1.JEI.23.2.023014 [52] Li D Z, Li X, Qin Y L, et al. Radar coincidence imaging: an instantaneous imaging technique with stochastic signals [J]. IEEE Transactions on Geoscience and Remote Sensing, 2014, 52(4): 2261-2277. doi: 10.1109/TGRS.2013.2258929 [53] Yu H. Research on sparse imaging algorithms for correlated imaging systems[D]. Hefei: University of Science and Technology of China, 2014. (in Chinese) [54] Yang H T, Wang K Z, Yuan B, et al. Microwave staring imaging based on range pulse compression and azimuth wavefront modulation[C]//10th European Conference on Synthetic Aperture Radar (EUSAR 2014), 2014: 1-4. [55] Yuan Y, Li C R, Li X H, et al. Sensitivity analysis on radiant performance of microwave intensity correlation image [J]. Remote Sensing Technology and Application, 2015, 1(30): 155-162. (in Chinese) [56] Zhou X, Wang H, Cheng Y, et al. Radar coincidence imaging for off-grid target using frequency-hopping waveforms [J]. International Journal of Antennas and Propagation, 2016, 2016: 1-16. doi: 10.1155/2016/8523143 [57] Zhou X L, Wang H Q, Cheng Y Q, et al. Radar coincidence imaging by exploiting the continuity of extended target [J]. IET Radar, Sonar & Navigation, 2017, 11(1): 60-69. [58] Zhou X, Wang H, Cheng Y, et al. Off-grid radar coincidence imaging based on variational sparse Bayesian learning [J]. Mathematical Problems in Engineering, 2016, 2016: 1782178. [59] Cao K C, Cheng Y Q, Liu K, et al. Coherent-detecting and incoherent-modulating microwave coincidence imaging with off-grid errors[J/OL]. IEEE Geoscience and Remote Sensing Letters(2021-11-13)https://ieeexplore.ieee.org/document/9612163. [60] Dai Q. Research on radar coincidence imaging technology in low SNR[D]. Changsha: National University of Defense Technology, 2014. (in Chinese) [61] Cao K C. Research on radar coincidence imaging with model mismatch[D]. Changsha: National University of Defense Technology, 2017. (in Chinese) [62] Yuan T Z. Research on radar imaging using electromagnetic vortex wave[D]. Changsha: National University of Defense Technology, 2017. (in Chinese) [63] Liu K. Study on the theory and method of electromagnetic vortex imaging[D]. Changsha: National University of Defense Technology, 2017. (in Chinese) [64] Laska J N, Wakin M B, Duarte M F, et al. A new compressive imaging camera architecture using optical-domain compression[C]//Conference on Computational Imaging IV, 2006: 606509. [65] Hunt J D. Metamaterials for computational imaging[D]. Durham: Duke University, 2013. [66] Duan P, Wang Y Y, Xu D G, et al. Single pixel imaging with tunable terahertz parametric oscillator [J]. Applied Optics, 2016, 55(13): 3670-3675. doi: 10.1364/AO.55.003670 [67] Chen S, Luo C G, Deng B, et al. Study on coding strategies for radar coded-aperture imaging in terahertz band [J]. Journal of Electronic Imaging, 2017, 26(5): 053022. [68] Gan F J, Luo C G, Liu X Y, et al. Fast terahertz coded-aperture imaging based on convolutional neural network [J]. Applied Sciences-Basel, 2020, 10(8): 2661. doi: 10.3390/app10082661 [69] Gan F J, Yuan Z Y, Luo C G, et al. Phaseless terahertz coded-aperture imaging based on deep generative neural network [J]. Remote Sensing, 2021, 13(671): 1-15. [70] Liu X Y, Wang H Q, Luo C G, et al. Terahertz coded-aperture imaging for moving targets based on incoherent detection array [J]. Applied Optics, 2021, 60(23): 6809-6817. doi: 10.1364/AO.428457 [71] Luo C G, Deng B, Wang H Q, et al. High-resolution terahertz coded-aperture imaging for near-field three-dimensional target [J]. Applied Optics, 2019, 58(12): 3293-3300. doi: 10.1364/AO.58.003293 [72] Yang H T, Zhang L J, Gao Y S, et al. Azimuth wavefront modulation using plasma lens array for microwave staring imaging[C]//IEEE Geoscience and Remote Sensing Symposium, 2015: 4276-4279. [73] Sleasman T, Boyarsk M, Imani M F, et al. Design considerations for a dynamic metamaterial aperture for computational imaging at microwave frequencies [J]. Journal of the Optical Society of America B, 2016, 33(6): 1098-1111. doi: 10.1364/JOSAB.33.001098 [74] Hunt J, Gollub J, Driscoll T, et al. Metamaterial microwave holographic imaging system [J]. Journal of the Optical Society of America A, 2014, 31(10): 2109-2119. doi: 10.1364/JOSAA.31.002109 [75] Gollub J N, Yurduseven O, Trofatter K P, et al. Large metasurface aperture for millimeter wave computational imaging at the human-scale [J]. Scientific Reports, 2017, 7: 42650. doi: 10.1038/srep42650 [76] Andreas P, Claire M W, Smith D R, et al. Enhanced resolution stripmap mode using dynamic metasurface antennas [J]. IEEE Transactions on Geoscience and Remote Sensing, 2017, 55(7): 3764-3772. doi: 10.1109/TGRS.2017.2679438 [77] Sleasman T, Boyarsky M, Pulido-Mancera L, et al. Experimental synthetic aperture radar with dynamic metasurfaces [J]. IEEE Transactions on Antennas and Propagation, 2017, 65(12): 6864-6877. doi: 10.1109/TAP.2017.2758797 [78] Cui T J, Wu R Y, Wu W, et al. Large-scale transmission-type multifunctional anisotropic coding metasurfaces in millimeter-wave frequencies [J]. Journal of Physics D:Applied Physics, 2017, 50(40): 404002. doi: 10.1088/1361-6463/aa85bd [79] Liu S, Cui T J. Concepts, working principles, and applications of coding and programmable metamaterials [J]. Advanced Optical Materials, 2017, 5(22): 1700624. doi: 10.1002/adom.201700624 [80] Wang L, Li L, Li Y, et al. Single-shot and single-sensor high/super-resolution microwave imaging based on metasurface [J]. Scientific Reports, 2016, 6: 26959. doi: 10.1038/srep26959 [81] Li Y B, Li L L, Xu B B, et al. Transmission-type 2-bit programmable metasurface for single-sensor and single-frequency microwave imaging [J]. Scientific Reports, 2016, 6: 23731. doi: 10.1038/srep23731 [82] Zhao M R, Zhu S T, Huang H L, et al. Frequency-polarization-sensitive metasurface antenna for coincidence imaging [J]. IEEE Antennas and Wireless Propagation Letters, 2021, 20(7): 1274-1278. doi: 10.1109/LAWP.2021.3077556 [83] Zhao M R, Zhu S T, Huang H L, et al. Frequency-diverse metamaterial cavity antenna for coincidence imaging [J]. IEEE Antennas and Wireless Propagation Letters, 2021, 20(6): 1103-1107. doi: 10.1109/LAWP.2021.3073679 [84] Chen S. Research on technology of three-dimensional terahertz coded-aperture imaging[D]. Changsha: National University of Defense Technology, 2018. (in Chinese) [85] Luo Z L. Research on coded aperture imaging based on programmable metasurface[D]. Changsha: National University of Defense Technology, 2018. (in Chinese)