针对近红外光谱高维小样本数据以及信号具有稀疏先验的特点,提出了小波变换(WT)结合正交匹配追踪(OMP)建立秦艽中龙胆苦苷定量分析模型的方法。将204个秦艽样品按照Kennard-Stone法以3∶1的比例进行划分,得到153个校正集样品和51个测试集样品,利用傅里叶变换近红外光谱仪采集样品的近红外光谱,采用WT对原始近红外光谱数据进行预处理,选择小波基函数为Daubechies(db5),小波分解层数为5,小波阈值为0.1,采用OMP建立龙胆苦苷的定量分析模型。结果表明:该模型预测性能较好,校正集对应的决定系数(R_C²)为0.994 0,校正均方根误差(RMSEC)为0.081 9,测试集对应的决定系数(R_P²)为0.985 4,预测均方根误差(RMSEP)为0.112 4;利用所建模型分析204个秦艽样品,龙胆苦苷预测值与参考值基本一致。

Aiming at the characteristics of sparse prior of high-dimensional small samples data and signals in near infrared spectroscopy, a method for the establishment of quantitative analysis model of gentiopicroside in Gentiana macrophylla by wavelet transform (WT) with orthogonal matching pursuit (OMP) was proposed. According to Kennard-Stone method, 204 Gentiana macrophylla samples were divided into 153 calibration set samples and 51 testing set samples at ratio of 3∶1. Fourier transform near infrared spectrometer was used to collect the near infrared spectra of samples, and WT was used to pretreat the original near infrared spectra data. The wavelet basis function was Daubechies (db5), the wavelet decomposition layer was 5, and the wavelet threshold value was 0.1. The OMP was used to establish the quantitative analysis model of gentiopicroside. As shown by the results, the prediction performance of this model was good. The determination coefficient of calibration set (R_C²) was 0.994 0, and root mean square error of calibration (RMSEC) was 0.081 9. The determination coefficient of testing set (R_P²) was 0.985 4, and root mean square error of prediction (RMSEP) was 0.112 4. The proposed model was used to analyze 204 Gentiana macrophylla samples, and the predicted values of gentiopicroside were basically consistent with the reference values.