基于近红外光谱和化学计量学的党参定性模型的建立

Discrimination of Codonopsis Pilosula According to Geographical Origin with Near Infrared Spectroscopy and Chemometrics

采用近红外光谱结合化学计量学,对来源于不同产地的81个党参样本进行快速、无损的定性研究。将党参样本粉碎后,采用光纤漫反射采集近红外光谱,并结合随机森林建立党参的定性模型,同时对比分析了不同光谱预处理方法(未处理、变量标准化+一阶导数、偏移校正+一阶导数)对鉴别结果的影响。结果显示,样本可按产地区分开;光谱经偏移校正+一阶导数处理后,训练集准确率达100%,测试集准确率为94%。研究表明,近红外光谱结合随机森林原理简单,易操作,准确率高,为作为快速鉴别党参的参考方法。

The combination of near infrared （NIR） spectrum technique with chemometrics provides a fast and nondestructive approach to study the Codonopsis pilosula in relation to its geographical origin. Near infrared spectrum were collected by using a fiber-optic in diffuse reflectance mode. Random Forests （RF） was applied for establishing classification model of Codonopsis pilosula samples （81） and different pretreatment methods （untreated, SNV-k 1st derivative and offset correctionq-1st derivative） were compared. After spectrum data were pretreated by offset correction q-＋ 1st derivative,the classification accuracy of random forests reached 100 % for training set and 94 M for test set. The study indicated that near infrared spectroscopy combined with random forests was simple,efficient and accurate, and could identify geographical origin of Codonopsis pilosula fast.