-
摘要: 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本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.Abstract: The ``small sample size'' (SSS) problem will cause the singularity and instability of the per class covariance matrices. This paper uses perturbing training samples to produce virtual training samples to overcome singularity of the per class covariance matrices. As a consequence, the classifier based on quadratic discriminant analysis (QDA) can be used directly in classification. The proposed QDA overcomes the problem that the parameters of regularized discriminant analysis (RDA) needs optimizing. Our experiments show that the QDA's recognition accuracy is superior to that of RDA if its parameters are optimized.
点击查看大图
计量
- 文章访问数: 2755
- HTML全文浏览量: 58
- PDF下载量: 1702
- 被引次数: 0