基于LLE-RF的中药三维荧光光谱分类识别

针对中药三维荧光光谱信号非线性问题,应用局部线性嵌入算法(LLE)对补虚类中药三维荧光光谱信号进行特征提取。使用荧光光谱仪测得不同浓度的24味补虚类中药的三维荧光光谱和等高线光谱图,并利用总体平均经验模态分解(EEMD)算法进行降噪处理,在此基础上,采用LLE算法对经过降噪处理后的三维荧光光谱数据进行特征提取; 然后,应用随机森林(RF)算法对提取的特征向量进行分类识别。为了验证LLE算法的有效性,将其与主成分分析(PCA)算法进行比较。实验结果表明: LLE-RF组合算法分类准确率达95%,比PCA-RF算法分类准确率高,从而验证了该算法的有效性。

Classification Recognition of Three Dimensional Fluorescence Spectrum Traditional Chinese Medicine Based on LLE-RF

To research the non-linearity of the three dimensional fluorescence spectrum signal of traditional Chinese medicine, locally linear embedding (LLE) algorithm was used to extract the feature of three dimensional fluorescence spectrum signals for tonifying herbs. On the basis of the three dimensional fluorescence spectrum and contour line spectrum of 24 traditional Chinese medicines of different concentrations measured by fluorescence spectrometer,and ensemble empirical mode decomposition (EEMD) algorithm for noise reduction, the LLE algorithm was used to extract the characteristics of the three dimensional fluorescence spectrum data after noisereduction.Then, the extracted features were classified and identified by using random forest (RF) algorithm.In order to verify the effectiveness of LLE algorithm, the principal component analysis (PCA) algorithm was compared. Experimental results showed that the classification accuracy of the LLE-RF combination algorithm was 95%, which was higher than PCA-RF algorithm which proved the effectiveness of the algorithm.