目的 基于特征性近红外光谱与化学计量学方法构建红芪搓条加工品的快速鉴别模型。方法 采用近红 外光谱仪获取 91 份红芪未搓条及搓条加工品的特征性近红外光谱信息；利用多元散射校正法（MSC）、标准正 态变量变换（SNV）、一阶导数（FD）、二阶导数（SD）、无平滑（No Smoothing）、Savitzky‑Golay 滤波（SG）及 Norris 导数滤波（ND）组合的方法处理原始近红外光谱。采用主成分分析-马氏距离（PCA-MD）、正交偏最小二乘判别 分析（OPLS-DA）及 VIP 值＞1 的 OPLS-DA 模型方法筛选搓条与未搓条红芪样品的差异性特征近红外光谱。 结果 搓条与未搓条加工红芪样品的近红外原始光谱差异性特征光谱位置不明显，需对光谱数据做进一步化学 计量学处理，以便准确判别搓条与未搓条红芪样品。光标法人工比对筛选出搓条与未搓条红芪样品的近红外光 谱具有差异性的波数范围为 4 317.02～7 577.39 cm-1 。PCA-MD 模型筛选出光谱预处理方法为 MSC+FD+SG （21∶3），搓条与未搓条红芪样品间存在 FD 光谱图上的差异，确定差异较大的波数范围为 4 535.75～4 913.73 与 5 434.42～6 124.81 cm-1 。OPLS-DA 模型结果显示，91 份红芪样品可按照搓条加工与未搓条加工明显分为 2 类，分别为野生、栽培、商品搓条样品一类，野生、栽培未搓条样品一类。人工比对选取的 FD 光谱波段 4 535.75～4 913.73 与 5 434.42～6 124.81 cm-1是区分搓条与未搓条红芪样品的差异波数范围。基于 VIP 值＞1 光谱波数建立的 OPLS-DA 模型可明显区分搓条与未搓条红芪样品，并确定区分搓条与未搓条红芪样品差异性 特征光谱波数 26 个。结论 本研究建立的基于特征性近红外光谱与化学计量学方法构建的红芪未搓条及搓条 加工品快速鉴别模型准确、可靠，可实现搓条与未搓条红芪样品的准确鉴别，可为搓条加工科学内涵的揭示提 供参考。

Objective To establish a rapid discrimination model for the rolled and non-rolled processed products of Hedysari Radix based on characteristic near-infrared （NIR） spectroscopy combined with chemometrics. Methods Characteristic NIR spectral data were acquired from 91 samples of non-rolled and rolled Hedysari Radix using a NIR spectrometer. The raw NIR spectra were processed using a combination of preprocessing methods， including multiplicative scatter correction （MSC），standard normal variate transformation （SNV），first derivative （FD），second derivative （SD）， No Smoothing， Savitzky-Golay （SG） filtering， and Norris derivative （ND） filtering. Principal component analysis-Mahalanobis distance （PCA-MD），orthogonal partial least squares-discriminant analysis （OPLSDA），and an OPLS-DA model based on VIP values ＞1 were employed to screen the differential characteristic NIR spectral bands between rolled and non-rolled samples. Results The raw NIR spectra showed no obvious characteristic positions for differentiation between rolled and non-rolled samples， necessitating further chemometric processing for accurate discrimination. Manual comparison using the cursor method preliminarily identified the differential wavenumber range of 4 317.02–7 577.39 cm⁻¹. The PCA-MD model，with optimal preprocessing identified as MSC+FD+SG （21∶3）， revealed differences in the FD spectra，pinpointing the wavenumber ranges of 4 535.75–4 913.73 cm⁻¹ and 5 434.42– 6 124.81 cm⁻¹ as significantly different. The OPLS-DA model results demonstrated clear separation of the 91 samples into two distinct clusters：one comprising wild，cultivated，and commercial rolled samples，and the other comprising wild and cultivated non-rolled samples. The FD spectral bands of 4 535.75–4 913.73 cm-1 and 5 434.42–6 124.81 cm-1 ， selected by manual comparison， were confirmed as key differential ranges. An OPLS-DA model built using VIP＞1 effectively discriminated the samples and identified 26 characteristic wavenumbers crucial for differentiation. Conclusion The rapid discrimination model established in this study，based on characteristic NIR spectroscopy and chemometrics， is accurate and reliable. It enables precise identification of rolled and non-rolled Hedysari Radix samples，providing a reference for elucidating the scientific connotation of the rolling processing technique.

R282.5

国家自然科学基金青年项目（82104345）；2025年全国老药工传承工作室建设项目（国中医药人教函〔2025〕181号）；甘肃省教育厅高校 教师创新基金项目（2025A-109，2023A-081，2026B-131）；甘肃中医药大学引进人才科研启动基金项目（2023YJRC-04）；甘肃省中药质量与 标准研究重点实验室 2024年度开放基金项目（ZYZL2024-03）。