FT-NIR and ATR-FTIR spectroscopy combined with machine learning for accurate identification of variants and hybrids of Gastrodia elata Blume

文献类型： 外文期刊

第一作者： Han, Duo

作者： Han, Duo;Wang, Yuanzhong;Han, Duo;Liu, Honggao

作者机构：

关键词： FT-NIR; ATR-FTIR; Machine learning; Gastrodia elata Blume; Variants

期刊名称：MICROCHEMICAL JOURNAL （ 影响因子：5.1； 五年影响因子：4.7 ）

ISSN： 0026-265X

年卷期： 2025 年 216 卷

页码：

收录情况： SCI

摘要： Different variants of Gastrodia elata Blume (GEB) have differences in chemical composition due to genetic characteristics, which inevitably affect their quality, so the identification of GEB is of great significance for the effective utilization of its resources. Unsupervised visualization models Principal Component Analysis and TDistributed Stochastic Neighbor Embedding (PCA and t-SNE), supervised models Partial Least Squares Discriminant Analysis and Support Vector Machine(PLS-DA and SVM), and a deep learning model Residual Convolutional Neural Network (ResNet) were constructed using FT-NIR, ATR-FTIR, and FT-NIR + ATR-FTIR dataset, respectively, to quickly and reliably identify Different species of GEB. The results showed that the visualization effect of t-SNE was better than PCA, but they could not classify four GEB. The performance of SVM model is better than PLS-DA model, but it requires complex preprocessing to improve the accuracy. The accuracy of ResNet model established by three spectral dataset reached 100 % in the training set, test set, and external validation set, it achieved the best effect without complex preprocessing. Comparing PLS-DA, SVM, and ResNet models under three spectral dataset, FT-NIR was relatively best. It can be inferred that ResNet model based on FT-NIR is more suitable for the identification of variants and hybrids of GEB. This method may be applied to GEB products and other research fields.

分类号：

  • 相关文献
作者其他论文 更多>>

微信小程序