特征与度量联合优化信息几何检测器

Information Geometry Detector Based on Joint Optimization of Feature and Metric

目前,信息几何检测器主要采用协方差矩阵特征模型,在矩阵流形上度量待检测单元数据与杂波数据间的差异,以区分目标与杂波,从而实现雷达目标检测.然而,在复杂杂波背景下,雷达回波信号信杂比低,杂波在其中占据主导地位.因此,含有目标回波的雷达回波信号与纯杂波具有统计相似性,该相似性使得二者在矩阵流形上较难区分,从而限制了信息几何检测器的性能优势.为突破特征表示所造成的性能增益限制,本文提出了基于特征与度量联合优化的信息几何检测器.首先设计了特征与度量可调的信息几何检测器灵活框架,并在此基础上基于纽曼-皮尔逊准则建立了特征与度量的联合优化模型,而后利用局部平坦假设与多层感知器,将联合优化模型中的决策变量参数化,并提出了双阶段优化求解方法.基于仿真数据与实测海杂波数据的实验结果表明,该方法检测性能优于现有信息几何检测器等典型目标检测方法,且在目标多普勒接近杂波谱峰时具有较大优势.

Nowadays,information geometry detectors mostly utilized covariance matrix model and measured the difference between the sample data from the cell under test and clutter data on the matrix manifold to distinguish them for radar target detection.However,under complex clutter backgrounds,the received signal with target echoes is clutter-dominated due to the low signal-to-clutter ratio,so the similarity between them in terms of statistics leaded to the unavailable distinguishability,so the performance advantage of information geometry detector was limited.To break through this limit,this paper proposed the information geometry detector based on a joint optimization of feature and metric.Specifically,this paper first designed the flexible framework of information geometry detector with a changeable signal feature and a metric.Then,on the basis of this framework,the Neyman-Pearson criterion based joint optimization with respect to feature and metric was established.By utilizing the locally flatness hypothesis and multilayer perceptron,the decision variables in the optimization problem were parameterized,and then the two-stage algorithm for this optimization problem was deduced.Based on the simulated data and real-recorded sea clutter data,the experiments show that the superiority of the proposed method than existing information geometry detectors and typical detection methods.Moreover,the experimental results demonstrate that the proposed method possesses the great advantage in slow moving target detection when the target Doppler closes to the peak of the clutter spectrum.