A Novel Pattern Recognition Method for Non-Destructive and Accurate Origin Identification of Food and Medicine Homologous Substances with Portable Near-Infrared Spectroscopy

In this study, a novel pattern recognition method named boosting-partial least squares-discriminant analysis (Boosting-PLS-DA) was developed for the non-destructive and accurate origin identification of food and medicine homologous substances (FMHSs). Taking <i>Gastrodia elata</i>, <i>Aurantii Fructus Immaturus</i>, and <i>Angelica dahurica</i> as examples, spectra of FMHSs from different origins were obtained by portable near-infrared (NIR) spectroscopy without destroying the samples. The identification models were developed with Boosting-PLS-DA, compared with principal component analysis (PCA) and partial least squares-discriminant analysis (PLS-DA) models. The model performances were evaluated using the validation set and an external validation set obtained one month later. The results showed that the Boosting-PLS-DA method can obtain the best results. For the analysis of <i>Aurantii Fructus Immaturus</i> and <i>Angelica dahurica</i>, 100% accuracies of the validation sets and external validation sets were obtained using Boosting-PLS-DA models. For the analysis of <i>Gastrodia elata</i>, Boosting-PLS-DA models showed significant improvements in external validation set accuracies compared to PLS-DA, reducing the risk of overfitting. Boosting-PLS-DA method combines the high robustness of ensemble learning with the strong discriminative capability of discriminant analysis. The generalizability will be further validated with a sufficiently large external validation set and more types of FMHSs.