-
远程亚纳秒脉冲激光目标回波信号具有幅度微弱、非平稳的特点。小波变换既保有博里叶变换方法的优点,又具有短时博里叶变换的良好局域特性。小波变换适合对信号的局部现象进行分析,在处理非平稳信号和微弱信号方面有着独特的优势。基于小波变换研究动态杂波背景下脉冲激光目标波形的检测,提出了适应背景动态杂波的波形提取算法。
在小波域,杂波背景下脉冲激光目标波形的检测按照Neyman-Pearson准则进行假设检验判决。设小波分解尺度为j,目标回波信号的小波系数可表示为:
式中:H1为有目标波形存在的假设;H0为无目标波形存在的假设;
$d_S^j $ 是尺度j下信号的小波系数;$d_C^j $ 是尺度j下杂波与噪声的小波系数。按照小波分解重构的要求,这里的小波系数通常是指小波分解得到的高频系数。由小波变换的理论可知,高斯噪声经过小波变换之后依然是高斯的。在杂波背景服从高斯分布的前提下,可以假设阳光背景光杂波与电路噪声的小波系数$d_C^j $ 服从均值为μ,方差为σ2的高斯分布。小波变换是线性变换,有:式中:Nj为目标波形小波系数均值的估计值。
在Neyman-Pearson准则下,小波域脉冲激光目标波形存在的似然比检验
${\rm{\Lambda (}}d_X^j {\rm{)}}$ 表示为:由似然比检验判决准则,有:
式中:调整参数λ可获得给定的虚警概率Pfa值。对上式求对数,有:
式中:
$\dfrac{{N}_{j}}{\sigma }$ 为信杂比。设统计量
$x = \dfrac{{d_X^j - \mu }}{\sigma }\sim N(0,1)$ ,由公式(6)可得到虚警概率为:这样,对于给定Pfa可由公式(11)得到阈值β。进一步,由公式(6)可得到检测概率Pd为:
在小波系数服从均值为μ,方差为σ2的高斯分布的前提下,得到尺度j下的小波分解系数的阈值为:
对信号进行小波分解,得到高频系数进dj行阈值处理,大部分高频系数可作为噪声置为零,再结合全部低频系数cj利用小波重构完成降噪。
对于高频系数进行置零,多少才是最合适的,结合不同评估方法形成软阈值、硬阈值等处理算法[10]。设高频系数dj经阈值处理后的小波系数为
$ {\hat d}_j $ ,按照软阈值方法,由硬阈值方法,有:
公式(10)中的统计量
$X = {{(d_X^j - \mu )} / \sigma }$ 是小波系数$d_X^j $ 对噪声小波系数的均值和标准差归一化的结果。由x~N(0,1),x与杂波均方根σ无关,而似然比判决值β由虚警概率Pfa确定。在β不变的情况下对x进行检测,可以获得恒定的虚警概率。在零均值高斯杂波背景下,这一目标波形检测提取方法具有较好的鲁棒性。实际上,阈值Thj与杂波背景的均值μ和方差σ2有关,均是未知,需要加以估计。采取删除措施[10],先将尺度j下的小波系数按绝对值大小进行排序,然后删除从最大值起始的一部分小波系数,认为这些被删除的小波系数是激光目标波形的小波系数,取剩余的小波系数作为杂波小波系数的估计
$ {\hat d}_j ,k(k = $ $ 1,2, \cdots , M_j )$ ,Mj为尺度j下总的小波系数个数。那么均值和标准差的无偏估计量分别为:如何选取这一部分最大的小波系数,可以采用Donoho阈值来确定。Donoho在小波变换的基础上提出了小波阈值去噪方法,并证明了该方法在最小方差意义下可达到最佳估计值。
式中:σj、n和Thj分别为尺度j下的小波系数的标准差、原始信号长度和阈值水平;med表示求中位数。通过Donoho阈值对小波系数进行处理,将最大的小波系数作为目标波形加以删除,剩余的小波系数采用公式(13)再进行阈值处理,小于该阈值的小波系数的数值置为零。要指出的是,公式(13)中的均值和标准差可由公式(16)和公式(17)加以估计,但阈值参数β是一个与回波波形幅值及杂波强度(信杂比)有关的参数。因此,通过实时估计杂波强度,动态调整阈值参数β,可以实现自适应地消减杂波,检测目标波形。所提算法步骤如下表1所示。
Input:Echo observation signal s(n) Output:Target waveform s’(n) Step1. For the target echo signal s(n) with clutter and noise, the symmetrical wavelet basis function is used for discrete wavelet transform, and the low-frequency coefficients cj,k and the high-frequency coefficients dj,k are obtained; Step2. The Donoho threshold Tj of wavelet decomposition high frequency coefficients is calculated,
and the wavelet coefficients larger than Tj are deleted from dj,k;Step3. According to the observed data, the root mean square of each segment of clutter is stored to evaluate the clutter intensity; Step4. According to the clutter condition, the threshold Thj is set by Eq. (13), and the coefficients less than the threshold value in the remaining wavelet decomposition high-frequency coefficients are set to zero, and $d_{j,k}' $ is obtained; Step5. cj,k, s’(n) is obtained by IDWT reconstruction of $d_{j,k}' $. Table 1. Adaptive detection algorithm of target echo waveform
Research on echo filtering algorithm of multi pulse laser range extended target in dynamic clutter background of airborne platform
doi: 10.3788/IRLA20200449
- Received Date: 2020-11-23
- Rev Recd Date: 2021-02-01
- Available Online: 2021-05-12
- Publish Date: 2021-03-15
-
Key words:
- airborne platform /
- dynamic target detection /
- extended target /
- target waveform image /
- wavelet transform
Abstract: Modern aircraft is equipped with airborne photoelectric detection system. Infrared thermal imager is used to search the azimuth of aircraft target in airspace, and pulsed laser rangefinder is used to measure the radial distance of target. Airborne pulsed laser target detection is a dynamic process. When the light spot moves on the target or the atmospheric turbulence refraction causes the echo beam to deviate from the receiving antenna, sometimes the target is absent, which makes it impossible to track the target stably. Only the amplitude information of target echo pulse is used to detect, which limits the effective range of laser target. In the dynamic clutter background of airborne platform, the problem of dynamic target detection can be solved by considering the pulse laser target as the range extended target and the echo signal as the target waveform image. In view of this, a multi pulse laser range extended target echo filtering algorithm based on wavelet transform was proposed. Experimental results show that the proposed algorithm can keep the waveform characteristics well.