论文部分内容阅读
小样本问题会造成各类协方差矩阵的奇异性和不稳定性.本文采用对训练样本进行扰动的方法来生成虚拟训练样本,利用这些虚拟训练样奉克服了各类协方差矩阵的奇异性问题,从而可以直接使用二次判别分析(Quadratic discriminant analysis,QDA)方法.本文方法克服了正则化判别分析(Regularized discriminant analysis,RDA)需要进行参数优化的问题.实验结果表明,QDA的模式识别率优于参数最优化时RDA算法的识别率.
The problem of small sample will lead to the singularity and instability of various covariance matrices.In this paper, the method of disturbing the training samples is used to generate the virtual training samples, and the virtual training samples are used to overcome the singularity problems of various covariance matrices , So that the Quadratic Discriminant Analysis (QDA) method can be used directly.This method overcomes the problem that the parameters need to be optimized for Regularized Discriminant Analysis (RDA) .The experimental results show that the pattern recognition rate of QDA is excellent Recognition rate of RDA algorithm in parameter optimization.