基于最小最大鉴别分析特征提取的雷达目标识别

针对基于雷达目标高分辨距离像的识别问题,本文提出了一种新的准则函数,并依据该准则函数采用神经网络来实现自适应的非线性特征提取、变换与识别。同时,由于准则函数只坜对特征提取后的类间散度和类内散度作出估计,因此在样本数相对较少的情况下也能得到较好的估计结果。该准则函数在形式上与Fisher准则函数相似,但它们之间存在着本质的差异。鉴于二者的形似之处,遂将基于该准则函数的特征提取方法冠以“最小最大鉴别分析特征提取”之名。将该方法用于五类缩比目标测量数据的特征提取与识别,取得了较好的识别效果。

Radar target identification based on minimum and maximum discriminant analysis feature extraction

A new criterion function for radar target identification is presented.According to this criterion function an adaptive nonlinear feature extraction algorithm was developed and compared with the discriminant analysis feature extraction (DAFE) method.Though this criterion is similar with Fisher Criterion in the form,they are widely divergent in essence.While for the similarity we still called this feature extraction method based on this criterion as "minimum and maximum discriminant analysis feature extraction (MMDAFE)".The experiment results show that this new method is practical.And by feature extraction and transform the identification probability is higher than using high resolution range profile directly.Perhaps,this is because there are some redundant information in the range profiles and feature extraction and reduction would remove the effects from the redundant information.