Research Status and Prospect of Human Movement Recognition Technique Using Through-Wall Radar
-
摘要: 在人体目标的动作识别应用中,穿墙雷达(TWR)具有隐蔽性高、探测能力强和不易受环境因素限制等优点,同时兼具良好的目标隐私信息保护能力,在武装反恐、安保监控和医疗看护等领域发挥出重要作用。为了梳理穿墙雷达对人体目标动作识别技术的发展脉络以及预测该技术的未来发展趋势,该文首先简要介绍穿墙探测的工作原理,并对不同体制穿墙雷达的特点进行比较和讨论;然后,围绕穿墙雷达人体动作识别应用中的雷达成像、特征参数提取和动作状态判决等关键技术,对国内外公开发表的相关文献进行了归纳分析;最后,对穿墙雷达的人体动作识别技术研究进行总结和展望,指出该技术在目前实际应用中所面临的潜在问题和挑战。Abstract: In applications of human action recognition, Through-Wall Radar (TWR) is a promising tool because of its outstanding advantages in aspects of concealment, detection ability and robustness against environmental restrictions. Besides, TWR can provide targets with satisfactory privacy protection. As a result, TWR is widely used in a series of areas including anti-terrorism, security monitoring and medical caring. To hackle and forecast the development process of the TWR-based human action recognition theory, the detection principle of different kinds of TWRs is first introduced in this article, and their properties are compared. Then aiming at the key technologies involved in human action recognition, such as radar imaging, feature information extraction, and action state judgement, the relevant literature published at home and abroad is classified and analyzed. Finally, the TWR-based human action recognition theory is summarized and prospected, and some potential problems and challenges in practical applications are pointed out.
-
1. 引言
随着对海洋资源开发、海洋权益维护等的日益重视,水声通信技术在军事和民用水下信息传输领域的重要性日益显现。但复杂多变的水声信道给信息的可靠传输带来了挑战,信号在高速率传输过程中会同时面临时变、严重码间干扰及衰落的影响[1]。为了解决此问题,通常考虑采用自适应均衡与分集技术。
自适应均衡是一种在接收机处进行自适应信号补偿的技术,可有效抑制码间干扰及噪声,矫正和补偿信道特性,从而可提高通信系统的可靠性。在水声通信中应用的传统自适应均衡算法主要有两种:最小均方误差(Least Mean Square, LMS)类自适应算法[2]和递归最小二乘(Recursive Least Square, RLS)类自适应算法[3]。相较于LMS,基于RLS类的自适应算法收敛性能更好,但其计算复杂度较高。而水声信道延迟扩展时间大,会导致严重的码间干扰,故传统自适应均衡算法由于其结构上的局限性,难以取得较好的误码率性能。近年来出现的深度学习算法,凭借其强大的学习能力和非线性拟合能力,成为了水声信道均衡算法研究的热点[4],给取得更好码间干扰消除效果带来了可能。文献[5]分别使用多层感知器(Multi-Layer Perceptron, MLP)和递归神经网络(Recursive Neural Network, RNN)实现了信道均衡,能获得比传统自适应均衡算法更好的误码率性能,但其仅能处理多径信道数很少的情况。文献[6]使用卷积递归神经网络(Convolutional Recursive Neural Network, CRNN)设计信道均衡器,可实现误码率性能的进一步改善。但CRNN存在局部最优解问题,且复杂度及训练成本较高,文献[7]考虑将5层深度神经网络(Deep Neural Network, DNN)应用到水声OFDM通信系统中,用于检测恢复发送端发送的信息。目前关于神经网络用于均衡方面的研究,还仅局限于单分支情况。
由于水下声波的衰落及传播特性,可考虑在水声通信中采用单发多收(Single Input Multiple Output, SIMO)分集技术,通过利用多个衰落相互独立的水声信道传送相同的信息,可更好地捕获声信号的能量及抵抗衰落的影响,改善水声通信链路的可靠性。在目前的分集技术中,主要的合并技术包括:等增益合并和最大比合并[8]。最大比合并性能最好,但其实现需已知对应各个分支的信道状态信息,而实际的水声环境下,换能器间的信道状态信息是难以获得的,因此,此方法不太适合于实际的水声通信系统。此外,在高速水声通信中,还存在码间干扰,因此不能仅采用合并技术处理,还需采用均衡技术,但如果多分支合并与均衡间相互独立,并不能充分发挥两者的联合性能,因此需考虑如何将多分支合并与均衡联合实现。文献[9]提出一种多分支合并与均衡联合实现算法,该自适应算法可基于LMS和RLS算法实现。文献[10]提出了一种基于归一化最小均方(Normalized Least Mean Square, NLMS)的多分支合并与均衡联合实现算法。文献[11]提出了一种基于均方误差准则的多分支合并与均衡算法,但其本质实现上仍然是各分支相互独立的。据文献调研可知,尽管在均衡领域,已有研究将智能学习算法用于均衡器,但还未有考虑基于智能学习方法将多分支合并和均衡联合实现的研究出现。
针对水声信道中严重衰落及码间干扰问题,本文提出一种基于深度学习的联合多分支合并与均衡算法。该算法基于DNN实现,可利用其强大学习能力和非线性拟合能力,更好的实现合并和消除码间干扰,在其实现中,多分支合并与均衡是联合实现的,可进一步提高联合性能。仿真结果验证了所提算法的可行性,也验证了其相对于已有方法的优势。
2. 系统模型
本文研究针对SIMO单载波水声通信系统,发射端有一个换能器,接收端有N个水听器,如图1所示。在图1中,h表示海水的深度,ht表示发射端换能器距海底的距离,hd(i)表示接收端第i个水听器序距海底距离,i∈{1,2,⋯,N},d表示换能器与水听器阵列的水平距离,hi(n)表示换能器与第i个水听器间水声信道的冲激响应,假设发射端换能器与接收端N个水听器间的水声信道hi(n)是相互独立的。
在此模型下,接收端第i个水听器的接收信号ri(n)可,其可表述为下式
ri(n)=s(n)∗hi(n)+ξi(n) (1) 其中i∈{1,2,⋯,N},s(n)表示独立等概率二进制相移键控(Binary Phase Shift Keying, BPSK)调制信号,s(n)∈{−1,1},∗表示卷积运算,ξi(n)表示零均值,方差为σ2i的加性高斯白噪声。
3. 基于深度学习的联合多分支合并与均衡
高速水声通信会同时面临水声通信中时变、严重码间干扰及衰落的影响,为了改善水声通信链路的可靠性,需要考虑同时解决。目前,已有合并均衡方法基于MMSE标准实现,深度学习方面则仅考虑了单分支情况,还未考虑多分支情况。本文提出了一种基于深度学习的联合多分支合并与均衡算法(Joint Multi-branch Merging and Equalization based on Deep Learning, JMME-DL),该算法与已有算法相比具有以下不同之处:(1)该算法基于深度学习网络实现了多分支均衡;(2)多分支合并和各分支均衡并非相互独立,而是联合实现的。
本文所提出JMME-DL算法的结构示意图如图2所示,该算法将直接处理各水听器接收到的信号,经如图2所示的深度学习网络处理后,通过判决即可恢复发送端发送的数据。
由图2可知,各分支网络和对各分支网络输出的合并并非相互独立,而是基于网络总输出计算误差进行联合更新的,每个分支会同时处理相应分支的若干组数据,第i个分支的第1层神经网络输入o(1)i可表示为
o(1)i=[ri(n),ri(n−1),⋯ri(n−Mi+1)ri(n+1),ri(n),⋯ri(n−Mi+2)⋮⋮⋱⋮ri(n+K−1),⋯⋯ ri(n−Mi+K)] (2) 其中,i∈{1,2,⋯,N},K表示每个分支1次处理的组数(这里假设各分支的组数均为相同的),Mi表示各个分支DNN网络的输入层神经元的数量。设各个分支的深度神经网络(DNN)的层数均为L,若第i个分支的第l层输出为o(l)i,则第l+1层的输出可通过下式计算得出
o(l+1)i=f(l)i(o(l)i)=fi(θ(l)(i)o(l)i+b(l)(i)) (3) 其中,θ(l)(i)代表第i个分支网络中第l层的权重矩阵,θ(l)(i)={θ(l)(i)j,k}, b(l)(i)代表第i个分支网络中第l层的偏置向量,b(l)(i)={b(l)(i)j},fi(⋅)表示SIGMOID激活函数,如图3所示,f(l)i(⋅)表示第l层的激活函数。网络层数增加时,每一层输入的分布会逐渐偏移,这会导致某些层的输入落入激活函数fi(⋅)的饱和区内,造成梯度消失,从而导致网络参数无法更新。为了解决梯度消失引起的“梯度弥散”问题,在图2中的每一层前均加入正则化(Batch Normalization, BN)层,BN层能将每一层的输入信息变换为服从均值为0,方差为1的标准正态分布的序列,信息经这样的正则化处理后,就可落入激活函数的敏感区域,梯度变化大,可加快网络的收敛速度。
由于采用的调制方式是BPSK调制,故可将其看作是一个二分类的问题。鉴于SIMOID激活函数的输出值取值范围为(0,1),这里将BPSK调制值取“+1”时定义为正向类“1”,取“–1”时定义为负向类“0”,阈值取0.5,在判决时可根据估计值与0.5的大小关系,判决恢复发送端发送的信息。由图2所示,第L−1层间的输出可表示为
o(L−1)i=f(L−1)(f(L−2)((⋯f(1)(o(1)i)))) (4) 最后总网络的输出˜stotal由各分支的输出加权得到˜stotal=f(n∑i(θ(L−1)(i)o(L−1)i+b(L−1)(i)))。考虑通信的目的是传递信息,期望接收端能正确恢复信息,故定义深度学习网络的损失函数为
Loss(θ,b)=1K‖s−˜stotal‖2 (5) 其中,s表示期望信息s=[s((n−1)K+1),⋯,s(nK)],n=1,2,⋯,˜stotal表示对s的估计值。从式(5)可以看出,损失函数取值越小,可以使得对信息的估计值越逼近于期望值,故可通过最小化损失函数求得最优化的网络参数(权重矩阵和偏置向量)。为了能使算法自适应于信道,可推导出令此损失函数最小的自适应算法。由随机梯度下降法,可得权重矩阵和偏置向量的更新公式为
θ(l)(i)=θ(l)(i)−α∂L∂θ(l)(i) (l=1,2,⋯,L−1) (6) b(l)(i)=b(l)(i)−α∂L∂b(l)(i) (l=1,2,⋯,L−1) (7) 其中,α表示学习率,其取值严重影响深度神经网络的收敛及稳态性能。根据反向传播算法,更新过程中的偏导部分可由下式计算得出
∂L∂θ(L−1)(i)=∂L∂o(L−1)i∂o(L−1)i∂θ(L−1)(i)=(o(L−2)i)T((s−˜stotal)⊙f′(n∑i(θ(L−1)(i)o(L−1)i+b(L−1)(i)))) (8) ∂L∂b(L−1)(i)=∂L∂o(L−1)i∂o(L−1)i∂b(L−1)(i)=sum{(s−˜stotal)⊙f′n∑i(θ(L−1)(i)o(L−1)i+b(L−1)(i))} (9) 其中,⊙表示哈达玛积,(⋅)T表示矩阵转置,sum{}表示对矩阵按列求和。由式(9),为了表述方便,定义神经网络第L−1层的误差项δL−1i为
δL−1i=∂L∂o(L−1)i=(s−˜stotal)⊙f′(n∑i(θ(L−1)(i)o(L−1)i+b(L−1)(i))) (10) 类比可推导出第l层的误差为
δli=∂L∂oli=(s−˜stotal)⊙f′(n∑i(θ(l)(i)o(l)i+b(l)(i))) (11) 由式(8)–式(11),对第i分支第l层的更新公式可表示为
∂L∂θ(l)(i)=(o(l−1)i)Tδli (l=1,2,⋯,L−1) (12) ∂L∂b(l)(i)=sum{δli} (l=1,2,⋯,L−1) (13) 将式(12)–式(13)代入式(6)–式(7),可得JMME-DL自适应算法
θ(l)(i)=θ(l)(i)−α(o(l−1)i)Tδli (l=1,⋯,L−1) (14) b(l)(i)=b(l)(i)−αδli (l=1,⋯,L−1) (15) 综上,所提JMME-DL算法实现过程如算法1所示。
表 1 JMME-DL算法输入:训练集:D={(ri(n),s(n))}Ni=1中K组数据;验证集: V;学习率:α;正则化系数:λ;迭代次数:M 初始化:θ,b repeat for i = 1 2 ··· M do (1) 从训练集D中选取K组数据样本 (2) 前馈计算,直到最后一层并计算总输出 (3) 反向传播计算每一层的误差 // 计算每一层参数的导数 ∂L∂θ(l)(i)=(o(l−1)i)Tδli (l=1,⋯,L−1) ∂L∂b(l)(i)=sum{δli} (l=1,⋯,L−1) // 更新参数 θ(l)(i)=θ(l)(i)−α(o(l−1)i)Tδli (l=1,⋯,L−1) b(l)(i)=b(l)(i)−αδli (l=1,⋯,L−1) until训练的模型在验证集V的错误率不再下降; 输出:θ,b 4. 仿真结果与分析
4.1 仿真条件
在仿真中,假设水声信道是半稳态的,即在一个数据包的发送过程中,信道是不变的,但对下一个数据包的传输来说,信道会发生变化。每个数据包由训练序列和数据序列构成。每个数据包的长度为1008个符号,其中前400个符号为训练序列,用于对网络的训练,后608个符号为数据信息,用于对网络的测试。在仿真中,JMME-DL各分支网络的输入神经元个数设定为16。为了在训练长度为400时得到更多的训练样本,各分支构成样本时,采用重复使用的方式,即第1~16个采样值构成第1个样本,2~17构成第2个样本,依次类推,400个训练符号可形成400组训练样本。换能器和水听器间的水声信道冲激响应基于文献[12]中的统计水声信道模型得出,所使用主要参数如表1所示,基于此水声信道模型可构建蒙特卡洛仿真,据表1中参数设置产生的水声信道冲激响应如图4所示。在仿真中深度学习网络的学习率α设置为0.005,正则化系数λ设置为0.8,迭代次数M设置为100。为了验证本文提出JMME-DL算法的有效性,与已有算法进行了误码率和收敛性能的对比,在仿真图中的误码率和收敛曲线分别以平均多个数据包下的相应曲线得出。
表 1 水声信道仿真主要参数仿真参数 数值 海水深度(m) 300 发射机深度(m) 100 水听器1深度(m)
水听器2深度(m)120
125水听器3深度(m) 130 发射机与水听器水平距离(m) 3000 水下传播系数 1.6 水下声速(m/s) 1500 载波频率(kHz) 10 带宽(kHz) 5 4.2 仿真结果与分析
本文选取对比算法如下:(1)文献[10]提出的基于NLMS的联合多分支合并与均衡算法(Joint Multi-branch Merging and Equalization based on NLMS, JMME-NLMS)。(2)文献[6]提出的基于卷积递归神经网络(CRNN)算法和等增益合并相结合的多分支处理算法,各分支是相互独立的,但对各分支的输出进行等增益合并,称之为基于CRNN的等增益合并多分支均衡算法(Equal-Gain Combing Multi-branch Equalization based CRNN, EGC-ME-CRNN)。(3)基于深度学习的单分支均衡算法(Single-branch Equalization based on Deep Learning, SE-DL)。(4)基于深度学习的均衡和等增益合并相结合的算法,各分支相互独立,仅对各分支的输出进行等增益合并,称之为基于DL的等增益合并多分支均衡算法(Equal-Gain Combing Multi-branch Equalization based on Deep Learning, EGC-ME-DL)。
深度神经网络的非线性拟合能力与网络层数具有一定关系,网络层数过少,非线性拟合能力较差;而网络层数过多时,可能导致过拟合,因此,深度神经网络的层数严重影响JMME-DL的性能。有鉴于此,首先通过仿真研究了网络层数对误码率性能的影响。在构建各分支深度神经网络时,假设不同分支的神经网络层数和结构相同。图5给出了网络层数对误码率性能的影响,图中分别给出了各分支网络层数为4层、5层、6层、7层和8层时的误码率曲线:4层网络结构为16-16-32-1;5层网络结构为16-16-24-32-1;6层网络结构为16-16-24-32-24-1;7层网络结构为16-16-24-32-24-16-1;8层网络结构为16-16-24-32-36-24-16-1。从图5可以看出,随着层数的增加,JMME-DL的误码率性能会有进一步的性能改善,但当层数达到一定程度时,再进一步增加层数时,误码率性能不会有较大的提升,反而会增加算法的复杂度。本文中在考虑网络层数的设定时,是考虑保证算法性能的同时,令网络尽可能轻量(计算复杂度低)。故综合考虑误码率性能和算法复杂度,选择各分支的网络层数为6会是一个较好的折衷。因此,在后续仿真中,设置各分支的网络层数为6。
图6给出了SE-DL, EGC-ME-DL, JMME-NLMS, EGC-ME-CRNN和所提JMME-DL算法间的误码率性能比较。从图6可以看出,EGC-ME-DL, JMME-NLMS, EGC-ME-CRNN和所提JMME-DL算法的误码率性能优于SE-DL。这是因为这几种算法都是多分支算法,相对于SE-DL来说,可以获得分集增益。从图6还可以看出,EGC-ME-DL, EGC-ME-CRNN和所提JMME-DL算法能获得比JMME-NLMS更好的误码率性能,这是因为这几种算法都是基于神经网络实现,其具有强大的非线性拟合能力,能更好地消除码间干扰。此外,我们还可以看出所提JMME-DL的误码率性能优于EGC-ME-DL、EGC-ME-CRNN算法。这是因为JMME-DL是基于深度学习网络实现的,且在其实现过程中,各分支网络和多分支的合并并非是相互独立的,而是联合实现的。
图7给出了不同算法的收敛性能比较,横轴表示深度学习算法的迭代次数。从图7中可以看出EGC-ME-DL和EGC-ME-CRNN的收敛速度接近,而所提JMME-DL算法能获得比EGC-ME-DL和EGC-ME-CRNN更好的收敛速度。此外,还可以从图7中看出,所提JMME-DL算法达到的损失函数值小于EGC-ME-DL和EGC-ME-CRNN算法,这与图6中的误码率性能是一致的,损失函数值越小,误码率性能越好,这反过来也证明了图6中结果的正确性。
5. 结束语
考虑更好地消除严重多径导致的码间干扰及抗水声信道衰落,本文提出一种基于深度学习的联合多分支合并与均衡算法。在该算法中,多分支合并和均衡并不是相互独立的,而是基于设计的深度学习网络联合实现的,因而能获得更好的合并与均衡性能。仿真结果表明,与已有算法相比,所提JMME-DL算法能借助深度学习网络的非线性拟合能力,更有效地消除码间干扰,从而获得更好的收敛及误码率性能。
-
图 2 常见障碍物材料对不同频率电磁波信号的衰减效用对比图[13]
图 4 Soldovieri等人[38]的人体目标探测场景和成像结果示意
图 5 Zhang等人[39]的人体目标探测场景和2维成像结果示意
图 6 Zhang等人[40]的人体目标探测场景和3维成像结果示意
图 7 Dubroca等人[48]的人体目标探测场景和成像结果示意
图 8 Gollub等人[49]的人体目标探测场景和成像结果示意
图 9 Wang等人[52]的人体目标探测场景和2维成像结果示意
图 10 Ahmad等人[53]的目标探测场景和3维成像结果示意
图 11 Kong等人[55]的人体目标探测场景和3维成像结果示意
图 12 Zhao等人[56]的人体目标探测场景和成像结果示意
图 13 Adib等人[57]的人体目标探测场景和成像结果示意
图 14 Chen等人[61]的目标特征参数提取结果示意
图 15 Kim等人[62]的目标探测场景与特征参数提取结果示意
图 16 Zeng等人[63]的目标探测场景与特征参数提取结果示意
图 17 Du等人[65]的目标特征参数提取结果示意
图 18 Orovic等人[68]的目标特征参数提取结果示意
表 1 部分穿墙雷达产品的性能参数
研发机构 产品名称 中心频率(GHz) 带宽(GHz) 最大探测距离(m) 距离分辨率(cm) 主要功能 时域公司(美国) Radar Vision 3.85 3.5 10 5 2维定位 劳伦斯 ⋅利物摩亚实验室(美国) MIR-I 2.5 1 50 15 2维定位 卡梅罗公司(以色列) Xaver 800 4.8 3.4 20 20 2维定位/3维成像 剑桥咨询公司(英国) Prism 200 1.95 0.5 20 30 3维成像 华诺星空(中国) CE200 ? ? 30 30 2维定位 必肯科技(中国) 警视-2 0.50 ? 9 ? 2维定位 凌天世纪(中国) YSR-120 1.2 1.2 12 ? 2维定位 表 2 不同体制穿墙雷达的探测特点比较
雷达体制 发射波形 优点 缺点 窄带
穿墙雷达单频/多频连续波信号 系统简单,抗静态干扰能力强,
信号处理速度快获取的目标信息量少,对目标参数的估计精度低,
识别准确率较差,能耗大超宽带
穿墙雷达窄脉冲信号 穿透能力强,分辨率高 存在探测范围和距离分辨率间的取舍矛盾,
抗干扰能力弱,探测存在盲区步进频/线性调频信号 同时获得优秀的探测范围和距离分辨率 抗干扰能力弱,信号处理实时性差,难以对快速变化的目标信息做出及时反应 伪随机码/噪声信号 穿透能力强,分辨率高,抗干扰能力强,隐蔽性强 发射信号的产生困难,系统成本高,功率受限于特定器件限制。信号的伪随机特性容易导致误差累积效应,
在长时间工作条件下性能不稳定 -
[1] 刘云, 薛盼盼, 李辉, 等. 基于深度学习的关节点行为识别综述[J]. 电子与信息学报, 2021, 43(6): 1789–1802. doi: 10.11999/JEIT200267LIU Yun, XUE Panpan, LI Hui, et al. A review of action recognition using joints based on deep learning[J]. Journal of Electronics &Information Technology, 2021, 43(6): 1789–1802. doi: 10.11999/JEIT200267 [2] 罗会兰, 童康, 孔繁胜. 基于深度学习的视频中人体动作识别进展综述[J]. 电子学报, 2019, 47(5): 1162–1173. doi: 10.3969/j.issn.0372-2112.2019.05.025LUO Huilan, TONG Kang, and KONG Fansheng. The progress of human action recognition in videos based on deep learning: A review[J]. Acta Electronica Sinica, 2019, 47(5): 1162–1173. doi: 10.3969/j.issn.0372-2112.2019.05.025 [3] 黄华宾. 遮蔽人体目标行为识别方法研究[D]. [硕士论文], 电子科技大学, 2019.HUANG Huabin. Research on method of hiden human targets behavior recognition[D]. [Master dissertation], University of Electronic Science and Technology of China, 2019. [4] CICCHETTI R, PISA S, PIUZZI E, et al. Numerical and experimental comparison among a new hybrid FT-music technique and existing algorithms for through-the-wall radar imaging[J]. IEEE Transactions on Microwave Theory and Techniques, 2021, 69(7): 3372–3387. doi: 10.1109/TMTT.2021.3061500 [5] 金添, 宋勇平. 穿墙雷达人体目标探测技术综述[J]. 电波科学学报, 2020, 35(4): 486–495. doi: 10.13443/j.cjors.2020040804JIN Tian and SONG Yongping. Review on human target detection using through-wall radar[J]. Chinese Journal of Radio Science, 2020, 35(4): 486–495. doi: 10.13443/j.cjors.2020040804 [6] 王宏. 超宽带穿墙雷达成像及多普勒特性研究[D]. [博士论文], 电子科技大学, 2010.WANG Hong. Study of UWB through-wall imaging and Doppler characteristic[D]. [Ph. D. dissertation], University of Electronic Science and Technology of China, 2010. [7] TAYLOR J D, 胡明春, 王建明, 孙俊, 等译. 超宽带雷达应用与设计[M]. 北京: 电子工业出版社, 2017: 328–340.TAYLOR J D, HU Mingchun, WANG Jianming, SUN Jun, et al. translation. Ultrawideband Radar: Applications and Design[M]. Beijing: Publishing House of Electronics Industry Press, 2017: 328–340. [8] GAUGUE A C and POLITANO J L. Overview of current technologies for through-the-wall surveillance[C]. SPIE 5989, Technologies for Optical Countermeasures II; Femtosecond Phenomena II; and Passive Millimetre-Wave and Terahertz Imaging II, Bruges, Belgium, 2005: 59891H. [9] BOREK S E. An overview of through the wall surveillance for homeland security[C]. 34th Applied Imagery and Pattern Recognition Workshop (AIPR'05), Washington, USA, 2006: 1–6. [10] 狄培. 便携式穿墙雷达成像技术研究[D]. [硕士论文], 西安电子科技大学, 2019.DI Pei. Research on portable through-wall radar imaging technology[D]. [Master dissertation], Xidian University, 2019. [11] 北京凌天智能装备集团股份有限公司. YSR-120 手持式穿墙雷达 [EB/OL]. https://www.bjltsj.com/ html/product/1004.html, 2015.BEIJING TOPSKY INTELLIGENT EQUIPMENT GROUP Co.,Ltd. YSR-120 Handheld, Through Wall Radar [EB/OL]. https://www.bjltsj.com/ html/product/1004.html, 2015. [12] 华诺星空股份有限公司. 军事反恐救援利器: 二维定位—CE系列穿墙雷达[EB/OL]. http://www.insight-safety.com/productinfo/243693.html, 2009. [13] 郭山红, 孙锦涛, 谢仁宏, 等. 电磁波穿透墙体的衰减特性[J]. 强激光与粒子束, 2009, 21(1): 113–117.GUO Shanhong, SUN Jintao, XIE Renhong, et al. Attenuation characteristics of electromagnetic wave penetrating walls[J]. High Power Laser and Particle Beams, 2009, 21(1): 113–117. [14] DING Yipeng, YU Xiali, ZHANG Juan, et al. Application of linear predictive coding and data fusion process for target tracking by Doppler through-wall radar[J]. IEEE Transactions on Microwave theory and Techniques, 2019, 67(3): 1244–1254. doi: 10.1109/TMTT.2018.2885973 [15] QI Fugui, LI Zhao, MA Yangyang, et al. Generalization of channel micro-Doppler capacity evaluation for improved finer-grained human activity classification using MIMO UWB radar[J]. IEEE Transactions on Microwave Theory and Techniques, 2021, 69(11): 4748–4761. doi: 10.1109/TMTT.2021.3076055 [16] TROMMEL R P, HARMANNY R I A, CIFOLA L, et al. Multi-target human gait classification using deep convolutional neural networks on micro-Doppler spectrograms[C]. 2016 European Radar Conference (EuRAD), London, UK, 2016: 81–84. [17] CHUMA E L and IANO Y. A movement detection system using continuous-wave Doppler radar sensor and convolutional neural network to detect cough and other gestures[J]. IEEE Sensors Journal, 2021, 21(3): 2921–2928. doi: 10.1109/JSEN.2020.3028494 [18] ZHENG Chen, XI Xiaoli, SONG Zhongguo, et al. Multilevel delay lock loop approach for wall clutter mitigation in through-the-wall radar imaging[J]. IET Microwaves, Antennas & Propagation, 2018, 12(12): 1986–1992. doi: 10.1049/iet-map.2018.5234 [19] SOLIMENE R and CUCCARO A. Front wall clutter rejection methods in TWI[J]. IEEE Geoscience and Remote Sensing Letters, 2014, 11(6): 1158–1162. doi: 10.1109/LGRS.2013.2288491 [20] YOON Y S and AMIN M G. Spatial filtering for wall-clutter mitigation in through-the-wall radar imaging[J]. IEEE Transactions on Geoscience and Remote Sensing, 2009, 47(9): 3192–3208. doi: 10.1109/TGRS.2009.2019728 [21] DING Yipeng, TANG Jingtian, XU Xuemei, et al. Echo interference suppression approach for Doppler through-wall radar[J]. IEEE Sensors Journal, 2015, 15(6): 3395–3402. doi: 10.1109/JSEN.2014.2374419 [22] 郑晨, 席晓莉, 宋忠国, 等. 基于子空间技术中奇异向量分析的穿墙雷达杂波抑制方法[J]. 电子学报, 2019, 47(4): 848–854. doi: 10.3969/j.issn.0372-2112.2019.04.012ZHENG Chen, XI Xiaoli, SONG Zhongguo, et al. A singular vector stationarity method for clutter mitigation in through-the-wall radar based on subspace method[J]. Acta Electronica Sinica, 2019, 47(4): 848–854. doi: 10.3969/j.issn.0372-2112.2019.04.012 [23] 李家强, 徐小敏, 卢宝宝, 等. 倾斜阵列下奇异值分解穿墙雷达杂波抑制[J]. 雷达科学与技术, 2019, 17(5): 531–537. doi: 10.3969/j.issn.1672-2337.2019.05.010LI Jiaqiang, XU Xiaomin, LU Baobao, et al. Clutter reduction of through-the-wall radar based on singular value decomposition under tilted array[J]. Radar Science and Technology, 2019, 17(5): 531–537. doi: 10.3969/j.issn.1672-2337.2019.05.010 [24] 赵中兴, 孔令讲, 贾勇, 等. 一种有效的穿墙雷达成像杂波抑制算法[J]. 雷达科学与技术, 2014, 12(1): 51–57. doi: 10.3969/j.issn.1672-2337.2014.01.009ZHAO Zhongxing, KONG Lingjiang, JIA Yong, et al. An Efficient clutter reduction method in through-wall-radar imaging[J]. Radar Science and Technology, 2014, 12(1): 51–57. doi: 10.3969/j.issn.1672-2337.2014.01.009 [25] 韩银萍, 刘丽, 王冰洁, 等. 基于鲁棒主成分分析的混沌穿墙成像雷达杂波抑制方法[J]. 电子器件, 2020, 43(1): 142–146. doi: 10.3969/j.issn.1005-9490.2020.01.029HAN Yinping, LIU Li, WANG Bingjie, et al. Clutter removal using robust principal component analysis for chaos through-wall imaging radar[J]. Chinese Journal of Electron Devices, 2020, 43(1): 142–146. doi: 10.3969/j.issn.1005-9490.2020.01.029 [26] 黄臣, 刘宏清, 罗臻, 等. 一种低秩联合稀疏模型下的杂波抑制方法[J]. 西安电子科技大学学报, 2019, 46(6): 60–66. doi: 10.19665/j.issn1001-2400.2019.06.009HUANG Chen, LIU Hongqing, LUO Zhen, et al. Method for suppressing clutters with the joint low rank and sparse model[J]. Journal of Xidian University, 2019, 46(6): 60–66. doi: 10.19665/j.issn1001-2400.2019.06.009 [27] WATERS A E, SANKARANARAYANAN A C, and BARANIUK R G. SpaRCS: Recovering low-rank and sparse matrices from compressive measurements[C]. The 24th International Conference on Neural Information Processing Systems (NIPS), Granada, Spain, 2011: 1089–1097. [28] 宋勇平, 金添, 陆必应, 等. 一种新的MIMO线阵穿墙成像模型及其环境参数估计[J]. 电子与信息学报, 2014, 36(12): 2980–2985. doi: 10.3724/SP.J.1146.2014.00018SONG Yongping, JIN Tian, LU Biying, et al. A novel linear MIMO array through-the-wall imaging model and its associated environmental parameters estimation[J]. Journal of Electronics &Information Technology, 2014, 36(12): 2980–2985. doi: 10.3724/SP.J.1146.2014.00018 [29] LEE D, FONG B, MORITA P, et al. Imaging of walking human behind the wall using impulse radar[C]. 2020 IEEE International Symposium on Antennas and Propagation and North American Radio Science Meeting, Montreal, Canada, 2020: 1263–1264. [30] RENEAU J and ADHAMI R R. Phase-coded LFMCW waveform analysis for short range measurement applications[C]. 2014 IEEE Aerospace Conference, Big Sky, USA, 2014: 1–6. [31] SHRAVANKUMAR S M, LENKA M K, and BANERJEE G. Integrated wideband frequency-hopping radar system for see-through-wall sensor application[C]. 2019 IEEE MTT-S International Microwave and RF Conference (IMARC), Mumbai, India, 2019: 1–7. [32] 夏正欢, 张群英, 叶盛波, 等. 一种便携式伪随机编码超宽带人体感知雷达设计[J]. 雷达学报, 2015, 4(5): 527–537. doi: 10.12000/JR15027XIA Zhenghuan, ZHANG Qunying, YE Shengbo, et al. Design of a handheld pseudo random coded UWB radar for human sensing[J]. Journal of Radars, 2015, 4(5): 527–537. doi: 10.12000/JR15027 [33] 王明阳. 穿墙雷达人体行为识别方法研究[D]. [博士论文], 电子科技大学, 2019.WANG Mingyang. Research on human action recognition exploiting through-wall radar[D]. [Ph. D. dissertation], University of Electronic Science and Technology of China, 2019. [34] 胡志鹏. 超宽带MIMO雷达系统设计与穿墙成像方法研究[D]. [硕士论文], 吉林大学, 2020.HU Zhipeng. Design of UWB MIMO radar system and research on the method of through-wall imaging[D]. [Master dissertation], Jilin University, 2020. [35] ANDRE D, WATSON F, and FINNIS M. Multistatic dual-polarimetric through-wall 3D-SAR[C]. 13th European Conference on Synthetic Aperture Radar (EUSAR 2021), 2021: 1–4. [36] GENNARELLI G and SOLDOVIERI F. A linear inverse scattering algorithm for radar imaging in multipath environments[J]. IEEE Geoscience and Remote Sensing Letters, 2013, 10(5): 1085–1089. doi: 10.1109/LGRS.2012.2230314 [37] FEDELI A, PASTORINO M, PONTI C, et al. Forward and inverse scattering models applied to through-wall imaging[C]. 2020 14th European Conference on Antennas and Propagation (EuCAP), Copenhagen, Denmark, 2020: 1–4. [38] SOLDOVIERI F, AHMAD F, and SOLIMENE R. Validation of microwave tomographic inverse scattering approach via through-the-wall experiments in semicontrolled conditions[J]. IEEE Geoscience and Remote Sensing Letters, 2011, 8(1): 123–127. doi: 10.1109/LGRS.2010.2051014 [39] ZHANG Wenji and HOORFAR A. Two-dimensional through-the-wall radar imaging with diffraction tomographic algorithm[C]. 2011 IEEE International Conference on Microwave Technology & Computational Electromagnetics, Beijing, China, 2011: 96–99. [40] ZHANG Wenji and HOORFAR A. Three-dimensional synthetic aperture radar imaging through multilayered walls[J]. IEEE Transactions on Antennas and Propagation, 2014, 62(1): 459–462. doi: 10.1109/TAP.2013.2287274 [41] 孟庆阳. 微波逆散射成像中的若干问题研究[D]. [博士论文], 浙江大学, 2017.MENG Qingyang. Several researches on microwave inverse scattering imaging[D]. [Ph. D. dissertation], Zhejiang University, 2017. [42] 蔡继亮, 童创明, 姬伟杰. 超宽带穿墙雷达成像技术研究现状[J]. 电讯技术, 2012, 52(9): 1541–1546. doi: 10.3969/j.issn.1001-893x.2012.09.025CAI Jiliang, TONG Chuangming, and JI Weijie. Recent progress of ultra-wideband through-the-wall radar imaging[J]. Telecommunication Engineering, 2012, 52(9): 1541–1546. doi: 10.3969/j.issn.1001-893x.2012.09.025 [43] SU Yurou and LU Guizhen. An improved inversion model for two-dimensional microwave imaging[C]. 2019 Cross Strait Quad-Regional Radio Science and Wireless Technology Conference (CSQRWC), Taiyuan, China, 2019: 1–3. [44] BOGERT B. Demonstration of delay distortion correction by time-reversal techniques[J]. IRE Transactions on Communications Systems, 1957, 5(3): 2–7. doi: 10.1109/TCOM.1957.1097511 [45] FINK M. Time reversal of ultrasonic fields. I. Basic principles[J]. IEEE Transactions on Ultrasonics, Ferroelectrics, and Frequency Control, 1992, 39(5): 555–566. doi: 10.1109/58.156174 [46] 王凌, 郑文军. 时间反转技术在超宽带穿墙雷达中的应用[J]. 无线电工程, 2012, 42(5): 42–44. doi: 10.3969/j.issn.1003-3106.2012.05.014WANG Ling and ZHENG Wenjun. Application of time reversal technique in ultra-wideband through-the-wall radar[J]. Radio Engineering, 2012, 42(5): 42–44. doi: 10.3969/j.issn.1003-3106.2012.05.014 [47] ODEDO V C, YAVUZ M E, COSTEN F, et al. Time reversal technique based on spatiotemporal windows for through the wall imaging[J]. IEEE Transactions on Antennas and Propagation, 2017, 65(6): 3065–3072. doi: 10.1109/TAP.2017.2696421 [48] DUBROCA R, FORTINO N, DAUVIGNAC J Y, et al. Time reversal-based processing for human targets detection in realistic through-the-wall scenarios[C]. 2011 8th European Radar Conference, Manchester, UK, 2011: 1–4. [49] 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 [50] KHORASHADI-ZADEH V and DEHMOLLAIAN M. Through a cinder block wall refocusing using SAR back projection method[J]. IEEE Transactions on Antennas and Propagation, 2019, 67(2): 1212–1222. doi: 10.1109/TAP.2018.2882599 [51] 杨泽民, 孙光才, 吴玉峰, 等. 一种新的基于极坐标格式的快速后向投影算法[J]. 电子与信息学报, 2014, 36(3): 537–544. doi: 10.3724/SP.J.1146.2013.00613YANG Zemin, SUN Guangcai, WU Yufeng, et al. A new fast back projection algorithm based on polar format algorithm[J]. Journal of Electronics &Information Technology, 2014, 36(3): 537–544. doi: 10.3724/SP.J.1146.2013.00613 [52] WANG Hong, NARAYANAN R M, and ZHOU Zheng’ou. Through-wall imaging of moving targets using UWB random noise radar[J]. IEEE Antennas and Wireless Propagation Letters, 2009, 8: 802–805. doi: 10.1109/LAWP.2009.2021586 [53] AHMAD F, ZHANG Yimin, and AMIN M G. Three-dimensional wideband beamforming for imaging through a single wall[J]. IEEE Geoscience and Remote Sensing Letters, 2008, 5(2): 176–179. doi: 10.1109/LGRS.2008.915742 [54] SONG Yongping, HU Jun, CHU Ning, et al. Building layout reconstruction in concealed human target sensing via UWB MIMO through-wall imaging radar[J]. IEEE Geoscience and Remote Sensing Letters, 2018, 15(8): 1199–1203. doi: 10.1109/LGRS.2018.2834501 [55] KONG Lingjiang, CUI Guolong, YANG Xiaobo, et al. Three-dimensional human imaging for through-the-wall radar[C]. 2009 IEEE Radar Conference, Pasadena, USA, 2009: 1–4. [56] ZHAO Dizhi, DAI Yongpeng, SONG Yongping, et al. A three-dimensional enhanced imaging method based on deconvolution[C]. 2017 3rd IEEE International Conference on Computer and Communications (ICCC), Chengdu, China, 2017: 2076–2079. [57] ADIB F, HSU C Y, MAO Hongzhi, et al. Capturing the human figure through a wall[J]. ACM Transactions on Graphics, 2015, 34(6): 219. doi: 10.1145/2816795.2818072 [58] WANG Mingyang, CUI Guolong, HUANG Huabin, et al. Through-wall human motion representation via autoencoder-self organized mapping network[C]. 2019 IEEE Radar Conference (RadarConf), Boston, USA, 2019: 1–6. [59] AN Qiang, WANG Shuoguang, YAO Lei, et al. RPCA-based high resolution through-the-wall human motion feature extraction and classification[J]. IEEE Sensors Journal, 2021, 21(17): 19058–19068. doi: 10.1109/JSEN.2021.3088122 [60] LAI C P and NARAYANAN R M. Through-wall imaging and characterization of human activity using ultrawideband (UWB) random noise radar[C]. SPIE 5778, Sensors, and Command, Control, Communications, and Intelligence (C3I) Technologies for Homeland Security and Homeland Defense IV, Orlando, USA, 2005: 186–195. [61] CHEN V C. Doppler signatures of radar backscattering from objects with micro-motions[J]. IET Signal Processing, 2008, 2(3): 291–300. doi: 10.1049/iet-spr:20070137 [62] KIM Y and LING Hao. Human. Human activity classification based on micro-Doppler signatures using a support vector machine[J]. IEEE Transactions on Geoscience and Remote Sensing, 2009, 47(5): 1328–1337. doi: 10.1109/TGRS.2009.2012849 [63] ZENG Zhengxin, AMIN M G, and SHAN Tao. Automatic arm motion recognition based on radar micro-Doppler signature envelopes[J]. IEEE Sensors Journal, 2020, 20(22): 13523–13532. doi: 10.1109/JSEN.2020.3004581 [64] RAM S S and LING Hao. Analysis of microDopplers from Human gait using reassigned joint time-frequency transform[J]. Electronic Letters, 2007, 43(23): 1309–1311. doi: 10.1049/el:20071515 [65] DU L, LI J, STOICA P P, et al. Doppler spectrogram analysis of human gait via iterative adaptive approach[J]. Electronics Letters, 2009, 45(3): 186–188. doi: 10.1049/el:20092769 [66] DING Yipeng, LIU Runjin, LI Zhengmin, et al. Human micro-Doppler frequency estimation using CESP-based Viterbi algorithm[J]. IEEE Geoscience and Remote Sensing Letters, 2021, 19: 3506105. [67] AMIN M G, AHMAD F, ZHANG Y D, et al. Human gait recognition with cane assistive device using quadratic time–frequency distributions[J]. IET Radar, Sonar & Navigation, 2015, 9(9): 1224–1230. doi: 10.1049/iet-rsn.2015.0119 [68] OROVIĆ I, STANKOVIĆ S, and AMIN M. A new approach for classification of human gait based on time-frequency feature representations[J]. Signal Processing, 2011, 91(6): 1448–1456. doi: 10.1016/j.sigpro.2010.08.013 [69] LI Po, WANG Dechun, and WANG Lu. Separation of micro-Doppler signals based on time frequency filter and Viterbi algorithm[J]. Signal, Image and Video Processing, 2013, 7(3): 593–605. doi: 10.1007/s11760-011-0263-3 [70] JOKANOVIĆ B and AMIN M G. Suitability of data representation domains in expressing human motion radar signals[J]. IEEE Geoscience and Remote Sensing Letters, 2017, 14(12): 2370–2374. doi: 10.1109/LGRS.2017.2765341 [71] GROOT S R, YAROVOY A G, HARMANNY R I A, et al. Model-based classification of human motion: Particle filtering applied to the micro-Doppler spectrum[C]. 2012 9th European Radar Conference, Amsterdam, Netherlands, 2012: 198–201. [72] LE H T, PHUNG S L, and BOUZERDOUM A. Human gait recognition with micro-Doppler radar and deep autoencoder[C]. 2018 24th International Conference on Pattern Recognition (ICPR), Beijing, China, 2018: 3347–3352. [73] LIU Liang, POPESCU M, SKUBIC M, et al. Automatic fall detection based on Doppler radar motion signature[C]. The 5th International Conference on Pervasive Computing Technologies for Healthcare (PervasiveHealth) and Workshops, Dublin, Ireland, 2011: 222–225. [74] FAIRCHILD D P and NARAYANAN R M. Classification of human motions using empirical mode decomposition of human micro-Doppler signatures[J]. IET Radar, Sonar & Navigation, 2014, 8(5): 425–434. doi: 10.1049/iet-rsn.2013.0165 [75] 蒋留兵, 李骢, 车俐. 利用二维小波包分解实现超宽带雷达人体动作识别[J]. 电子测量与仪器学报, 2018, 32(8): 69–75. doi: 10.13382/j.jemi.2018.08.010JIANG Liubing, LI Cong, and CHE Li. Human motion recognition using ultra-wide band radar based on two-dimensional wavelet packet decomposition[J]. Journal of Electronic Measurement and Instrumentation, 2018, 32(8): 69–75. doi: 10.13382/j.jemi.2018.08.010 [76] DING Yipeng, LEI Chengxi, XU Xuemei, et al. Human micro-Doppler frequency estimation approach for Doppler radar[J]. IEEE Access, 2018, 6: 6149–6159. doi: 10.1109/ACCESS.2018.2793277 [77] EROL B, AMIN M G, and BOASHASH B. Range-Doppler radar sensor fusion for fall detection[C]. 2017 IEEE Radar Conference (RadarConf), Seattle, USA, 2017: 819–824. [78] KILIÇ A, BABAOĞLU I, BABALIK A, et al. Through-wall radar classification of human posture using convolutional neural networks[J]. International Journal of Antennas and Propagation, 2019, 2019: 7541814. doi: 10.1155/2019/7541814 [79] ZHANG Yang, QI Fugui, LV Hao, et al. Bioradar technology: Recent research and advancements[J]. IEEE Microwave Magazine, 2019, 20(8): 58–73. doi: 10.1109/MMM.2019.2915491 [80] CORTES C and VAPNIK V. Support-vector networks[J]. Machine Learning, 1995, 20(3): 273–297. doi: 10.1023/A:1022627411411 [81] YANG Yinan, ZHANG Wenxue, and LU Chao. Classify human motions using micro-Doppler radar[C]. SPIE 6944, Biometric Technology for Human Identification V, Orlando, USA, 2008: 69440V. [82] LI Jingli, PHUNG S L, TIVIVE F H C, et al. Automatic classification of human motions using Doppler radar[C]. The 2012 International Joint Conference on Neural Networks (IJCNN), Brisbane, Australia, 2012: 1–6. [83] LI Fangmin, YANG Chao, XIA Yuqing, et al. An Adaptive S-method to analyze micro-Doppler signals for human activity classification[J]. Sensors, 2017, 17(12): 2769. doi: 10.3390/s17122769 [84] ALVEE B I, TISHA S N, and CHAKRABARTY A. Application of machine learning classifiers for predicting human activity[C]. 2021 IEEE International Conference on Industry 4.0, Artificial Intelligence, and Communications Technology (IAICT), Bandung, Indonesia, 2021: 39–44. [85] SMITH G E, WOODBRIDGE K, and BAKER C J. Naive Bayesian radar micro-Doppler recognition[C]. 2008 International Conference on Radar, Adelaide, Australia, 2008: 111–116. [86] RICCI R and BALLERI A. Recognition of humans based on radar micro-Doppler shape spectrum features[J]. IET Radar, Sonar & Navigation, 2015, 9(9): 1216–1223. doi: 10.1049/iet-rsn.2014.0551 [87] PADAR M O, ERTAN A E, and CANDAN Ç Ĝ. Classification of human motion using radar micro-Doppler signatures with hidden Markov models[C]. IEEE Radar Conference, Philadelphia, USA, 2016: 1–6. [88] LEI Peng, WANG Jun, GUO Peng, et al. Automatic classification of radar targets with micro-motions using entropy segmentation and time-frequency features[J]. AEU - International Journal of Electronics and Communications, 2011, 65(10): 806–813. doi: 10.1016/j.aeue.2011.01.013 [89] 谢非, 徐贵力, 吕东岳, 等. 一种基于决策树的多种人体姿态识别方法[P]. 中国专利, 101533467B, 2013.XIE Fei, XU Guili, LV Dongyue, et al. Method for identifying a plurality of human postures based on decision tree[P]. China Patent, 101533467B, 2013. [90] KIM Y and MOON T. Human detection and activity classification based on micro-Doppler signatures using deep convolutional neural networks[J]. IEEE Geoscience and Remote Sensing Letters, 2016, 13(1): 8–12. doi: 10.1109/LGRS.2015.2491329 [91] KIM Y and TOOMAJIAN B. Hand gesture recognition using micro-Doppler signatures with convolutional neural network[J]. IEEE Access, 2016, 4: 7125–7130. doi: 10.1109/ACCESS.2016.2617282 [92] PAPANASTASIOU V S, TROMMEL R P, HARMANNY R I A, et al. Deep Learning-based identification of human gait by radar micro-Doppler measurements[C]. 2020 17th European Radar Conference (EuRAD), Utrecht, Netherlands, 2021: 49–52. [93] ZHU Jianping, CHEN Haiquan, and YE Wenbin. A hybrid CNN–LSTM network for the classification of human activities based on micro-Doppler radar[J]. IEEE Access, 2020, 8: 24713–24720. doi: 10.1109/ACCESS.2020.2971064 [94] ZHANG Zhenyuan, TIAN Zengshan, and ZHOU Mu. Latern: Dynamic continuous hand gesture recognition using FMCW radar sensor[J]. IEEE Sensors Journal, 2018, 18(8): 3278–3289. doi: 10.1109/JSEN.2018.2808688 [95] WANG Saiwen, SONG Jie, LIEN J, et al. Interacting with soli: Exploring fine-grained dynamic gesture recognition in the radio-frequency spectrum[C]. The 29th Annual Symposium on User Interface Software and Technology, Tokyo, Japan, 2016: 851–860. -