Rapid and accurate classification of Dioscorea species using attenuated total reflection-fourier transform infrared (ATR-FTIR) spectroscopy integrated with chemometric analysis and convolutional neural network (CNN)

Background Due to its high market demand and nutritional value, Chinese yam ( Dioscorea oppositifolia ) is often adulterated with Dioscorea hamiltonii, Dioscorea fordii, and Dioscorea alata . Discriminating among Dioscorea species is challenging because of their morphological similarities. Hence, this research employed Attenuated Total Reflection-Fourier Transform Infrared (ATR-FTIR) spectroscopy combined with chemometric analysis and a Convolutional Neural Network (CNN) to discriminate Dioscorea species. Method A total of 107 Dioscorea samples from China were collected, dried, ground, and sieved. ATR-FTIR spectra were acquired and preprocessed through ATR correction, smoothing, baseline correction, normalisation, and arithmetic processing. Chemometric techniques, including Principal Component Analysis (PCA), PCA-Class, and Orthogonal Partial Least Squares Discriminant Analysis (OPLS-DA), were applied to the spectra. Results and discussion ATR-FTIR spectra revealed significant absorption peaks; however, PCA could not clearly differentiate the Dioscorea species. Compared with PCA-Class, OPLS-DA demonstrated higher accuracy (95.24%), sensitivity (90.48%), and specificity (96.83%) in discriminating Dioscorea species. The low RMSEE, RMSECV, and RMSEP values (