SAR目标多尺度概率密度估计与识别

针对SAR目标识别问题,提出了一种基于核非线性映射的SAR目标多尺度概率密度特征的估计方法,并利用该特征进行目标识别。首先将SAR目标图像在多尺度域中分解,按一定规则建立多尺度根矢量;将多尺度根矢量经非线性映射到另一空间中,在该空间中利用基于核函数的技术结合parzen窗非参数估计得到概率密度函数。通过这一途径得到的多尺度概率密度分布挖掘了目标散射在尺度之间的相互关系,分布特征之间的相对熵测度可以用与目标分类识别。以MSTAR实测SAR目标数据集进行了多尺度概率密度估计和目标识别试验和分析,试验结果表明了提出方法的有效性。

Estimation of multi-scale probability density for SAR targets and its application in ATR

A method to estimate multi-scale probability density for SAR targets based on kernel non-linear mapping is proposed,and its application for SAR ATR is investigated.Firstly,SAR target imagery is decomposed in multi-scale domain,then multi-scale vectors are constructed according to a fixed rule.The multi-scale vectors are mapped using kernel-based technique into another space in which the multi-scale probability density is estimated using a parzen window.The multi-scale probability characteristic exploits the relationship of cross-scale of SAR target scattering.ATR is implemented by cross-entropy measurement between multi-scale probabilities.The experiment results are given using publicly released SAR data from DARPA's MSTAR program.