Deep learning-based identification of drying methods and quality prediction of Dendrobium officinale

Abstract

Dendrobium officinale (D. officinale) possesses numerous active compounds with significant medicinal and nutritional potential. This work evaluates the potential of Fourier transform infrared (FTIR) and near infrared (NIR) spectroscopy combined with partial least squares discriminant analysis (PLS-DA), back-propagation neural network (BPNN), support vector machine (SVM), and residual convolutional neural network (ResNet) in identifying and classifying D. officinale from different drying methods. Furthermore, the Long Short-Term Memory Net (LSTM) and dry matter content (DMC) were employed to quantify the D. officinale DMC across various drying modalities. The deep learning model showed superior classification performance compared to the conventional machine learning model. Both the training and test datasets showed 100% accuracy for the ResNet model built using FTIR and NIR spectra. The Ratio of Performance Deviations (RPD) values of the LSTM regression models developed from raw FTIR and NIR spectra exceed 2. This work holds considerable practical importance for assessing the quality of D. officinale, optimizing processing, and ensuring the quality of market supply.