基于均值似然估计的激光探测微动特征提取和分离  

Extraction and Separation of Micro-Motion Feature Based on Mean Likelihood Estimation in Laser Detection

作　　者：郭力仁 胡以华[1] 王云鹏[1] 

机构地区：[1]解放军电子工程学院脉冲功率激光技术国家重点实验室,电子制约技术安徽省重点实验室,安徽合肥230037

出　　处：《光学学报》2017年第4期180-190,共11页Acta Optica Sinica

基　　金：国家自然科学基金(61271353);安徽省自然科学基金(1308085QF123)

摘　　要：最大似然估计是提取目标微动特征参数的最佳估计方法,但直接用网格法求解计算量巨大,且激光探测微多普勒回波信号对应的代价函数具有高度非线性,存在多个局部最大值。为此,提出均值似然估计与蒙特卡罗结合的估计方法,给出了最大似然参数估计的闭合表达式,再通过设计压缩似然函数获得全局最大值,通过蒙特卡罗法抽样并计算循环均值估计出参数。该方法避免了传统方法中对高精度初始值和复杂迭代算法的依赖,能够实现参数的联合估计。对于多分量微多普勒信号,该方法可在参数估计的同时实现各微动分量分离,且不增加算法的复杂性。对仿真和实验数据进行估计,结果表明,该方法在达到近似于最大似然估计性能的同时可有效降低计算复杂度并确保了全局收敛,实现信号的分离和参数估计。Maximum likelihood estimation （MLE） is the optimal estimator for target micro-motion feature parameter extracting. However, the grid search will cause the enormous computational amount, and the cost function of laser detection of micro-Doppler echo signals has high nonlinearity and exists many local maxima. A new method combining the mean likelihood estimation and the Monte Carlo method is proposed to solve this problem. A closed-form expression of maximum likelihood parameter estimation is derived. Then the compressed likelihood function is designed to obtain the global maximum. The parameters are estimated by Monte Carlo method sampling and calculating the circular mean value. The dependence of hight accurate initial values and the complex iteration algorithms are avoided in this method, and the joint estimation of parameters can be realized. Furthermore, for multi-component micro-Doppler signal, the presented algorithm can separate the micro-motion component signals at the same time with the estimations, which will not add complexity of algorithm. Applied to the simulated and experimental data, the proposed method achieves similar performance as MLE with less computational complexity. Meanwhile, this method guarantees the global convergence and realizes signals separation and parameters estimation.

关 键 词：测量 激光雷达 微多普勒 参数估计 信号分离 

分 类 号：TN95[电子电信—信号与信息处理]