基于分步相关成分分析的中药材质量鉴别神经元分类器

A Neural Classifier for Identifying the Quality of Chinese Medicinal Materials Based on Stepwise Correlative Components Analysis

提出并构建了一种基于分步相关成分分析的神经元分类器 ( SCCA-HBP) ,并将其用于中药材质量模式分类 .通过从色谱分析所得到的高维数据集中分步提取分类相关成分 ,获取化学模式特征向量 ,使神经元分类器输入模式向量的维数降低 .此外 ,提出用带输出误差死区的混合 BP算法训练神经元分类器 ,提高了网络学习训练速度和分类准确性 .以 3 2个当归样品质量等级分类鉴别为例考察本方法 ,分类正确率为1 0 0 % ,优于 PCA-BP( 84.4% )和 SCCA-BP( 90 .6% )方法 ;且训练时间仅为 BP算法的 5 4.2 % .

A neural classifier based on stepwise correlative component analysis, named SCCA-HBP, for classifying the quality pattern of Chinese Medicinal Materials(CMM) is proposed. The chemical pattern features are extracted by stepwise acquirement of the class correlative components from chromatographic analysis dataset with a high dimension, and then are used as the inputs in the neural classifier to reduce the dimension of input variables. Further, a hybrid BP algorithm with dead interval of error is derived for training the neural classifier in order to increase training speed and classification accuracy. The performance of the neural classifier is tested by using a set of 32 Angelica samples with different quality grades. The classification accuracy of SCCA-HBP is 100%, better than PCA-BP( 84.4%) and SCCA-BP(90.6%). Moreover, the training time of HBP is 54.2% of the cost obtained with BP algorithm.