Data fusion of FT-NIR and ATR-FTIR spectra for accurate authentication of geographical indications for Gastrodia elata Blume

Chuanmao Zheng,Jieqing Li,Honggao Liu,Yuanzhong Wang
DOI: https://doi.org/10.1016/j.fbio.2023.103308
IF: 5.318
2023-11-04
Food Bioscience
Abstract:Gastrodia elata Blume ( G. elata Bl.), with its excellent nutritional and medicinal value from Zhaotong, has been protected by geographical indication (GI). Accurate certification of its origin is a prerequisite to safeguard consumer interests and maintain the market. Four different regions and three varieties of G. elata Bl. from Zhaotong were used in this study (n = 262). Tri-step infrared spectroscopy was used for ATR-FTIR spectral analysis to filter out fingerprint regions for data fusion with FT-NIR spectra, after which conventional discriminant models (PLS-DA and GS-SVM) were built. The second derivative (SD), multiple scattering correction (MSC), and Savitzky-Golay (SG) preprocessing were also performed on the spectra, and it was found that the preprocessing improved the performance of the PLS-DA model. The optimal model results in GS-SVM, based on mid-level data fusion of principal components (PCs) and latent variables (LVs), with sensitivity, specificity, accuracy, and Matthews correlation coefficient (MCC) of 1 for the test set. Furthermore, the residual convolutional neural network (ResNet) models were built, based on FT-NIR full spectra, band 3600−2700 cm −1 (MFA) and band 1750−500 cm −1 (MFB). Their accuracy in both train and test sets exceeds 97%, and the loss function curve is close to 0, which indicates that these three bands can be used as a fingerprint area to verify the GI of Zhaotong G. elata Bl. This study provides a fast, non-invasive method for the authentication of food or medicinal plant GI.
food science & technology
What problem does this paper attempt to address?