基于最大散度差判别分析的一种目标识别方法  

Method of Target Recognition Based on Maximum Scatter-difference Discriminant Analysis

作　　者：张善文[1] 李萍[1] 井荣枝[1] 张云龙[1] 

机构地区：[1]郑州大学西亚斯国际学院,郑州451150

出　　处：《系统仿真学报》2013年第3期441-444,共4页Journal of System Simulation

基　　金：河南省教育厅科学技术研究重点项目(12B120012);河南省重大科技攻关计划项目(122102210429);西亚斯国际学院引进人才项目(2012YJRC01;2012YJRC02)

摘　　要：针对线性判别分析(LDA)在多类高维小样本模式的分类中存在的"小样本问题"和"次优性问题",提出了一种基于最大散度差判别准则的监督维数约简方法。首先,构造类内和类间离散度函数;然后采用最大散度差判别准则设计最佳判别目标函数,得到映射矩阵和提取分类特征。该方法省略了求解逆矩阵过程,从而避免了传统的LDA存在的小样本问题;最后,在真实飞机图像数据库上的识别实验结果验证了该算法的有效性。Aiming at the Small-Sample-Size （SSS） and the ＂inferior＂ problems in linear discriminant analysis （LDA） in the classification of multi-class high-dimensionality pattern, based on maximum scatter-difference criterion, a supervised method of dimension reduction was proposed. Firstly, the within-class and between-class scatters were constructed. Then, the maximum scatter-difference criterion was adopted to design the optimal discriminant objective function. The projection matrix was obtained, by which the classification features were extracted. The SSS problem was utterly avoided by omitting to solve the converse matrix. Finally, the recognition experiments on the real-world aircraft image database verify the effectiveness of the proposed method.

关 键 词：飞机目标识别 线性判别分析 最大散度差判别分析 小样本问题 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]