Physical fingerprint transformation of herbal medicines powders using near-infrared spectroscopy

Abstract

To achieve the rapid determination of physical parameters of herbal medicine. A method based on near-infrared (NIR) spectroscopy was proposed. The potential of direct standardization, partial least squares regression and generalized regression neural network (GRNN) for physical fingerprint transformation of Paeoniae Radix Alba, Scutellariae Radix, Sinomenii Caulis, and Pueraria Lobatae Radix, were investigated. The results revealed that the predictive capacity of GRNN models was the best. Except for a few parameters in the validation samples of Pueraria Lobatae Radix and Sinomenii Caulis, the mean absolute deviations of all other physical parameters were less than 0.5. The similarity between the actual and predicted physical fingerprints of all validation samples and test samples was high when using the GRNN models (cosine coefficient > 0.99, Mahalanobis distance < 1.30). Additionally, the simplified GRNN models based on the 40 selected variables of hygroscopicity and angle of repose still showed the ideal predictive capacity.