目的 探讨近红外光谱（near infrared spectroscopy，NIRS）技术在天麻Gastrodia elata质量评价中的应用，通过对不同蒸制程度的天麻进行判别分析，进一步构建天麻有效成分含量预测模型，为天麻质量评价提供新方法。方法 建立天麻中天麻素（gastrodin，GAS）、对羟基苯甲醇（p-hydroxybenzyl alcohol，HBA）、巴利森苷B（parishin B，PB）、巴利森苷C（parishin C，PC）、巴利森苷A（parishin A，PA）含量测定的高效液相色谱法，并以其测定值为参比。采集天麻样品的NIRS，结合线性判别分析（linear discriminant analysis，LDA）算法，建立不同蒸制时间的天麻定性判别模型。选择偏最小二乘法（partial least squares，PLS）等化学计量学方法建立有效成分的测定值与NIRS的定量校正模型，对建模过程的各个阶段进行优化，构建天麻各有效成分的最优PLS定量模型。结果 基于LDA算法建立的天麻蒸制程度的定性判别模型准确度达到96.2%，混淆矩阵图与接收器工作特性（receiver operating characteristic，ROC）曲线评价模型的预测性能良好；天麻近红外原始光谱经标准正态变换（standard normal variate，SNV）或平滑法（savitzky-golay，SG）预处理后，以竞争性自适应重加权抽样-偏最小二乘法（competitive adaptive eeweighted sampling-partial least squares，CARS-PLS）构建的定量模型准确度较高，建立的GAS、HBA、PB、PC、PA最佳PLS定量模型的校正决定系数（R²c）分别为0.975 3、0.986 4、0.970 0、0.963 6、0.965 9，预测决定系数（R²p）分别为0.970 4、0.984 0、0.977 9、0.978 6、0.985 5，5个定量模型的预测偏差（residual prediction deviation，RPD）均大于6。表明NIRS定量模型预测值与测定值具有良好的线性关系，模型预测效果良好。结论 所建立的天麻近红外LDA定性和CARS-PLS定量模型准确、可靠，可实现天麻蒸制程度的定性鉴别以及GAS、HBA、PB、PC、PA 5个有效成分含量的快速定量分析，为天麻的质量评价与控制提供新的参考。

Objective To explore the application of near infrared spectroscopy (NIRS) technology in the quality evaluation of Tianma (Gastrodiae Rhizoma), a qualitative analysis was conducted to distinguish the degree of steaming of Gastrodiae Rhizoma. Furthermore, a prediction model was established to determine the contents of active components, aiming to provide a new method for quality evaluation of Gastrodiae Rhizoma. Methods A high-performance liquid chromatography (HPLC) method was established to measure the contents of gastrodin (GAS), p-hydroxybenzyl alcohol (HBA), parishin B (PB), parishin C (PC) and parishin A (PA) in Gastrodiae Rhizoma, which were used as the reference value. The near infrared spectroscopy of Gastrodiae Rhizoma with different degrees of steaming were collected. Linear discriminant analysis (LDA) was used to establish the discrimination model of the degree of steaming of Gastrodiae Rhizoma. The quantitative calibration model between the near infrared spectrum and the contents of active components to be measured was established by partial least squares (PLS) and other chemometrics methods. Each part of the modeling process was optimized respectively to construct the optimal PLS quantitative model for the active ingredients of Gastrodiae Rhizoma. Results The accuracy of the discrimination model based on LDA algorithm reached 96.2%, and the performance of model evaluated by the confusion matrix diagram and ROC curve was good. After pretreatment by standard normal variate (SNV) or Savitzky-Golay (SG), the quantitative analysis model constructed by competitive adaptive reweighted sampling - partial least squares regression (competitive adaptive reweighted sampling-partial least squares, CARS-PLS) had high accuracy. The correction determination coefficient (R²c) of GAS, HBA, PB, PC and PA models was 0.975 3, 0.986 4, 0.970 0, 0.963 6, and 0.965 9; The prediction determination coefficient (R²p) was 0.970 4, 0.984 0, 0.977 9, 0.978 6, and 0.985 5. The residual prediction deviation (RPD) values for all five quantitative models exceeded 6. The predicted values of NIRS models and the measured values of HPLC showed a good linear relation, which presented a great prediction ability of the models. Conclusion The established NIR-LDA qualitative model and CARS-PLS quantitative model were accurate and reliable, effectively identifying the steaming degree of Gastrodiae Rhizoma and nondestructively determining the contents of active components. These findings provide a novel reference for quality evaluation and control during the process of Gastrodiae Rhizoma.

R286.2

特色炮制技术规律发掘——煨制（GZY-KJS-2022-050）;江苏省研究生科研创新计划项目（KYCX23-2023）