近红外光谱技术结合偏最小二乘判别分析检测三七品质

Combination of near-infrared spectroscopy and partial least squares discriminant analysis in detecting the quality of Panax notoginseng

中文关键词:  三七  近红外光谱  品质鉴定  偏最小二乘判别分析  竞争自适应重加权采样

英文关键词:Panax notoginseng  near-infrared spectroscopy  quality discriminate  partial least squares discriminant analysis  competitive adaptive reweighted sampling

中文摘要:

      目的 通过近红外光谱技术实现不同等级三七样品的快速鉴别。方法 采集等级A (20头)、等级B (30头)、等级C (40头)、等级D (60头)4种不同等级三七样品的近红外光谱, 构建偏最小二乘判别分析(partial least squares discriminant analysis, PLS-DA)分类器模型鉴别4种等级的三七样品, 同时为了减近红外光谱中的冗余波长变量, 进一步优化模型的判别结果, 利用竞争自适应重加权采样(competitive adaptive reweighted sampling, CARS)算法提取近红外光谱中的特征变量。结果 所构建的PLS-DA分类器模型对等级C和等级D的三七样品, 鉴别准确率达到100%, 但是对于等级A和等级B的三七样品因为存在误判, 鉴别准确率仅为0%和20%。经过CARS算法提取近红外光谱特征变量后, 光谱变量数大幅减少, 从1557个变量下降到78个变量。以优选后的特征变量构建的CARS-PLS-DA分类器模型更加简化, 对4种等级三七样品的预测均方根误差均明显下降, 说明模型的预测分类变量更接近真实的分类变量, 鉴别结果更加准确。同时, 对4种等级三七样品的鉴别准确率显著上升, 其中对于等级C和等级D的鉴别准确率为100%, 对于等级B的鉴别准确率从20%提升到100%, 等级A鉴别准确率从0%提升到75%。结论 所构建的CARS-PLS-DA分类器模型对4种等级的三七样品具有更好的鉴别效果, 可以实现不同等级三七的品质鉴定。

英文摘要:

      Objective To realize the rapid identification of different grades of Panax notoginseng samples by near-infrared spectroscopy technology. Methods Collecting the near infrared spectroscopy of 4 kinds of different grades of Panax notoginseng, including grade A (20 tou), grade B (30 tou), grade C (40 tou), grade D (60 tou), partial least squares discriminant analysis (PLS-DA) classifier model was used to rapid discriminate the quality of Panax notoginseng. In order to reduce redundant wavelength variables of near infrared spectroscopy and optimize the discriminant results of model, competitive adaptive reweighted sampling (CARS) was used to extract characteristic wavelength variables of the near infrared spectroscopy. Results The constructed PLS-DA classifier model could be used to rapid discriminate the Panax notoginseng grade of C and grade D, with discriminant accuracy were both 100%, However, the discriminant accuracy was only 0% and 20% for the Panax notoginsenggrade of grade A and grade B, because misjudgment was found. And the number of characteristic variables were reduced from 1557 to 78 by CARS. And then, the CARS-PLS-DA classifier model was built by those characteristic variables. As a result, the CARS-PLS-DA classifier model was more simple, and the root mean square errors of prediction of different grades of Panax notoginseng were decreased obviously, indicating that the prediction classification variables of the model were closer to the real classification variables and the identification results were more accurate. Besides, the discriminant accuracy of Panax notoginseng of different grades increased significantly, among which the discriminant accuracy of grade C and grade D were 100%, the discriminant accuracy of grade B increased from 20% to 100%, and the discriminant accuracy of grade A increased from 0% to 75%. Conclusion The CARS-PLS-DA classifier model has better identification effect on Panax notoginseng,of different grades, and can realize the quality identification of Panax notoginseng of different grades.