Rapid and non-destructive identification of Gastrodia elata Blume geographical origin by infrared spectroscopy combined with chemometrics

Gastrodia elata Blume has high medicinal value and is also used as a cooking ingredient for daily health care. It is widely distributed, and identifying its geographical origin is important for guiding market supervision and meeting consumer needs. This study used near-infrared (NIR), attenuated total reflection-Fourier transform infrared (ATR-FTIR) spectroscopy and a data fusion strategy to identify the origins of Gastrodia elata Blume. Initially, two unsupervised learning methods, PCA and t-SNE, were utilised to visualise and analyse Gastrodia elata Blume from seven origins. The findings showed that it was not possible to distinguish between samples from different origins using the visualisation of analytical outcomes. Afterwards, through the comparison of multiple preprocessing methods, the PLS-DA model established by NIR data using FD+SNV preprocessing, and the SVM model that was established by NIR+ATR-FTIR data using SD preprocessing, the accuracy reached 100% (training sets) and 98.41% (test sets), respectively. Furthermore, a residual convolutional neural network (ResNet) model was built using two-dimensional correlation spectroscopy (2DCOS) images, which obtained 100% accuracy using NIR and NIR+ATR-FTIR data, respectively. In contrast, the established ResNet model obtains a robust model without preprocessing steps, providing a rapid, nondestructive and accurate method for origins identification of Gastrodia elata Blume. • IR spectroscopy captured region-specific chemical signatures in G. elata Bl. • Compared the effects of different spectral preprocessing methods on the modeling results. • The ResNet model demonstrates outstanding performance in geographical origin identification. • High robustness and performance even when the sample is unbalanced.