基于多分辨率特征的UWB SAR二维广义似然比目标检测方法

A 2-D GLR Target Detection Approach of UWB SAR Based on Multi-resolution Feature

针对超宽带合成孔径雷达(UWB SAR)工作体制及探测背景的特殊性,结合成像过程,分析了不同分辨率条件下目标与杂波的区别,在此基础上采用了两种形式的一阶自回归(AR)模型实现对多分辨率序列的建模,并提出了广义似然比(GLR)的二维计算方法。该方法大大提高了多分辨率特征提取的稳定性,基于实际UWB SAR图像的试验结果表明:利用多分辨率特征可明显增强图像信杂比,从而提高UWB SAR目标检测效果。

There are many features used to distinguish target from clutter in Synthetic Aperture Radar(SAR) target detection,such as amplitude feature,polarimetric feature,azimuthal feature,multi-resolution feature.There are many reports about the first three features,but there are very few reports about the development of multi-resolution feature.The approaches proposed in concerned references are effective to improve the performance of SAR target detection.But most of them discuss the multi-resolution feature for target detection of high-frequency SAR,so the proposed approaches are commonly suitable for the target detection of high-frequency SAR.Ultra-Wide Band Synthetic Aperture Radar(UWB SAR) can be used to detect the concealed targets because it works at low-frequency,and the corresponding detection background is the strong clutter produced by trunks.The application of multiresolution feature in UWB SAR target detection are analyzed,and the approaches suitable for UWB SAR target detection are proposed.In this paper,we establish the equivalent models of target and trunk clutter in UWB SAR images according to electromagnetic scattering theory based on the particularity of UWB SAR operation system.The differences between target and trunk clutter under different multiresolution are analyzed from UWB SAR image.The analysis supplies a key basis for the extraction of multiresolution feature in UWB SAR images.Two forms of first-order Auto-Regression(AR) model are used to deal with the multiresolution sequences.In the first AR model,we discuss its statistic distribution of residual to represent the differences between target and trunk clutter.In the second AR model,we discuss its statistic distribution of coefficient to represent the differences between target and trunk clutter.In two forms of first-order AR model,the corresponding definitions of Generalized Likelihood Ratios(GLR) are given.The definition of 2-D GLR is proposed based on two forms of AR model.The performance of the 2-D GLR is more robust in the multiresolution feature extraction because it integrates two forms of first-order AR model.The three steps of 2-D GLR calculation based on UWB SAR image are given: 1) generating multiresolution image sequences,2) training statistic model,3) calculating 2-D GLR.The multiresolution feature extraction experiment is accomplished in an actual UWB SAR image for the two 1-D GLRs and the 2-D GLR proposed in this paper.The results of the experiment show that the multiresolution features corresponding to the proposed three GLRs can all be used to improve the signal-clutter ratio(SCR) of the original image effectively,and the performance of the 2-D GLR is better than the two 1-D GLRs.