基于改进的子类判决分析的SAR目标特征提取与识别

SAR Target Feature Extraction and Recognition Based on Improved Clustering-based Discriminant Analysis

针对大多文献中假设合成孔径雷达(SAR)数据服从单模分布带来的问题,该文提出改进的子类判决分析(ICDA),它假设SAR目标数据服从更合理更实际的多模分布。首先采用快速全局k-均值聚类算法找到每类目标的子类划分,然后基于子类判决分析(CDA)准则寻找最优的投影矢量,使得投影后不同类别的子类样本之间距离最大而每个子类内部的样本散布最小。用美国运动和静止目标获取与识别(MSTAR)计划录取的SAR地面静止目标数据的实验结果表明,ICDA可获得较好的对真实目标的分类性能和对干扰目标的拒判能力。

In many literatures, Synthetic Aperture Radar （SAR） data is usually supposed to obey the unimodal distribution, unsuitable in the applications. To overcome the limitation, an Improved Clustering-based Discriminant Analysis （ICDA） method is proposed, which assumes the distribution of each class for SAR data is multimodal, a more reasonable and practical assumption. The detailed procedure of ICDA is to first partition each class of the SAR data into multiple clusters via the fast global k-means clustering algorithm, and then try to find the projection vectors such that the projections of every pair of clusters from different classes are well separated while the within-cluster scatter is minimized. Experimental results performing on SAR ground stationary targets based the Moving and Stationary Target Acquisition and Recognition （MSTAR） public database show that ICDA has better classification capabilities of three true objects classes and rejection capabilities of two confusers classes.