沈国芳 黄俊航 许麦成 金强

摘要 目的：建立一种基于近红外（NIR）高光谱成像技术融合近红外光谱和图像纹理特征，鉴别不同产地红参药材的方法。方法：提取红参ROI近红外高光谱数据，采用多种预处理算法对光谱数据进行降噪处理。利用灰度共生矩阵（GLCM）和灰度游程矩阵（GLRLM）提取图像纹理特征，实现NIR光谱和图像纹理数据融合。利用偏最小二乘判别分析（PLS-DA）和支持向量机分类（SVC）建立产地分类模型。结果：全波段光谱融合GLRLM所构建的模型性能最佳，准确率分别为90.0%和91.2%。进一步地使用混淆矩阵和ROC曲线对模型进行评估。混淆矩阵中SVC模型表现优异，对吉林、黑龙江和辽宁3个产地的分类准确率可达100%、91%和83%;经ROC特征曲线评估，2个模型的最优曲线下面积值分别达到了0.97和0.96。结论：本研究为快速鉴别红参药材不同产地提供了一种新方法。

关键词 高光谱成像;红参;数据融合;产地鉴别

Discriminant Analysis of Red Ginseng from Different Origins Based on NIR-Hyperspectral Imaging Technology

SHEN Guofang1，HUANG Junhang2，XU Maicheng3，JIN Qiang3

（1 Hangzhou Institute for Food and Drug Control，Hangzhou 310022，China; 2 College of Pharmaceutical Science，Zhejiang University，Hangzhou 310058，China; 3 Hangzhou Huqingyutang Pharmaceutical Co.，Ltd.，Hangzhou 311100，China）

Abstract Objective：To establish a method based on near-infrared（NIR） hyperspectral imaging technology which could identify red ginseng from different origins by fusing NIR spectra and image texture features.Methods：The near-infrared hyperspectral data of red ginseng ROI were extracted to be further de-noised by various pre-processing algorithm.Texture features are extracted from images using gray co-occurrence matrix（GLCM） and gray run matrix（GLRLM），and near-infrared spectroscopy and image data are fused.Partial least squares discriminant analysis（PLS-DA） and support vector machine classification（SVC） were used here to establish the origin classification model.Results：The model constructed by full-band spectral combined with GLRLM gained the best performance，with accuracy of 90.0% and 91.2%，respectively.The model was further evaluated using confusion matrix and ROC curves.The SVC model performed better in the confusion matrix，with the classification accuracy of 100%，91% and 83% for red ginseng from Jilin，Heilongjiang and Liaoning.According to ROC characteristic curve evaluation，the areas under the optimal curve of the two models are 0.97 and 0.96，respectively.Conclusion：This research provides a new method for rapid identification of red ginseng from different origins.

Keywords Hyperspectral imaging; Red ginseng; Data fusion; Origin identification

中圖分类号：R282.5文献标识码：Adoi：10.3969/j.issn.1673-7202.2021.23.003

红参为五加科植物人参Panax ginseng C.A.Mey.的栽培品经蒸制后的干燥根和根茎。在我国主产于东北3省，具有大补元气，复脉固脱，益气摄血的功效[1]。红参作为一种常用中药材，在中药制剂中应用广泛，随着对人参需求量的提升，人参栽培受限于连作障碍问题，其产地由吉林为主向东北各地扩展，不同产地的人参及其加工品红参存在着较大的质量差异，为确保中药制剂质量稳定，加强红参原料质量控制，对红参产地进行鉴别区分具有较大意义[2-6]。

传统红参鉴别以经验判断真伪优劣……