Abstract Poisonous Illicium lanceolatum A. C. Smith was found mixed with Illicium verum in the postharvest stage, which was difficult to separate the two species with naked eyes, especially broken samples and fallen follicles. A Vis/NIR hyperspectral imaging (HSI) system with wavelength range between 400 and 1000 nm was used as a non-destructive method to distinguish the two species with the purpose of safety control for Illicium verum. Firstly, spectral comparison and explanatory PCA was conducted to prove the discriminability between the two species in either face-up or back-up placement position, and a good clustering effect for I. lanceolatum samples was discovered. Then, linear partial least squares discriminant analysis (PLSDA) and nonlinear support vector machine (SVM) were developed to classify the two species based on full-wavelength, key wavelengths selected by successive projections algorithm (SPA) and regression coefficients (RC) and the most contributing principal components, in which SPA-PLSDA was selected as the most effective model (CCRc = 100%, CCRv = 96.88% and CCRcv = 98.44%). Furthermore, three forms (i.e., intact, broken and fallen follicle) of samples with different arrangements and two placements (face-up and back-up) were used to verify the applicability of the enhanced model. Correct classification rate (CCR) of the three independent tests were satisfactory (CCRset1 = 98.44%, CCRset2 = 100% and CCR set3 = 95.83%), and the sensitivity of I. verum (≥91.67%) and I. lanceolatum (100%) also proved that all I. lanceolatum could be identified using HSI technique, which establishes a theoretical and modelling foundation for the development of high-throughput online sorting equipment in future.

中文翻译：

---

使用 Vis/NIR 高光谱成像结合化学计量学无损区分八角茴香和有毒掺杂物

摘要 有毒八角花AC Smith在收获后与八角花混合，肉眼难以区分，尤其是破碎的样品和脱落的毛囊。使用波长范围在 400 到 1000 nm 之间的 Vis/NIR 高光谱成像 (HSI) 系统作为一种非破坏性方法来区分这两种物种，目的是对八角进行安全控制。首先，通过光谱比较和解释性PCA证明了两种物种在正面或背面放置位置之间的可区分性，并发现了对I. lanceolatum样品的良好聚类效果。然后，开发了线性偏最小二乘判别分析 (PLSDA) 和非线性支持向量机 (SVM) 以基于全波长对这两个物种进行分类，连续投影算法（SPA）和回归系数（RC）选择的关键波长和贡献最大的主成分，其中SPA-PLSDA被选为最有效的模型（CCRc = 100%, CCRv = 96.88% and CCRcv = 98.44% ）。此外，使用具有不同排列和两种放置方式（正面朝上和背面朝上）的三种形式（即完整、破损和掉落的卵泡）的样本来验证增强模型的适用性。三个独立测试的正确分类率（CCR）令人满意（CCRset1 = 98.44%，CCRset2 = 100% 和 CCR set3 = 95.83%），以及 I. verum (≥91.67%) 和 I. lanceolatum (100%) ) 也证明了所有的 I. lanceolatum 都可以使用 HSI 技术进行识别，