以96批栀子不同炮制品为研究对象, 高效液相色谱测定栀子苷含量为参考值, 利用近红外光谱仪积分球漫反射测定其光谱图, 建模波段取8 660~7 500, 6 650~5 600和4 900~4 000 cm-1, 以标准正态变换（SNV）和二阶导数法（2nd derivative）为预处理方法, 主成分数为8, 采用偏最小二乘法（PLS）对83批栀子样品建立栀子苷的定量校正模型, 最终以13批栀子不同炮制品对模型进行验证。 结果, 定量模型的内部交叉验证决定系数（R2）为0992 85, 校正均方差（RMSEC）为0240, 预测均方差（RMSEP）为0254, 内部交叉验证均方差（RMSECV）为0386 91, RMSEP/RMSEC=106。 模型验证得到的相对分析误差（RPD）为881, 绝对偏差范围-039%~023%, 说明模型预测性较好。 通过相关系数法, 优选样品装样量、 扫描次数、 重复次数、 分辨率实验条件; 并由近红外一阶导和二阶导图, 除去温湿度和样品水分影响波段, 结合栀子苷对照品近红外光谱图, 确定建模波段。 首次利用NIRS法建立栀子不同炮制品栀子苷定量校正模型, 方法简单快速, 模型稳定可靠、 准确性高, 可同时应用于不同炮制品栀子中栀子苷含量的预测。

In the study, 96 samples of Gardenia jasminoids Ellis in different preparation methods were collected, including model building with 83 samples and validation with 13 samples. The reference analysis was performed using a HPLC method to determine the content of geniposide. Method for the quantification of geniposide in Gardenia jasminoids Ellis in different preparation methods with NIRS are studied by integral sphere of diffuse reflection. The spectral regions from 8 660 to 7 500 cm-1, from 6 650 to 5 600 cm-1 and from 4 900 to 4 000 cm-1 were selected for the calculation of the quantitative model. The spectral data of geniposide were processed using the second derivative and standard normal variate transformation (SNV). The quantitative model of geniposide was built based on partial least squares (PLS) method with main components (8) with 83 samples of Gardenia jasminoids Ellis. Finally, validation of the quantitative model was accomplished with 13 samples of Gardenia jasminoids Ellis in different preparation methods. The correlation coefficients(R2), the root meat square error of calibration(RMSEC), the root meat square error of prediction(RMSEP), the root meat square error of cross validation(RMSECV) and RMSEP/RMSEC of the calibration model in geniposide were 0992 85, 0240, 0254, 0386 91 and 106. The ratio of standard deviation of the validation set to standard error of prediction (RPD) from model validation was 881, the range of absolute deviation was -039%~023%. The model established has good predictability. Besides, experimental conditions also have impact on the testing results, including scan frequency, amount of measured sample, the number of replications and resolution determined by correlation coefficient method. What’s more, using first derivative spectra, second derivative spectra and near-infrared spectra of geniposide to confirm the spectral regions can distinct the spectral regions according to temperature differences humidity and moisture content. For the first time, a quantitative model with NIRS is established for rapid determination of geniposide in Gardenia jasminoids Ellis with different preparation methods. The quantitative model of geniposide was stable and reliable, which can rapidly and accurately determine the content of geniposide in Gardenia jasminoids Ellis with different preparation methods with NIRS simultaneously.