Detection of Adulteration of Ziziphi Spinosae Semen Based on Near Infrared Hyperspectral Imaging

Abstract

Abstract Based on the near-infrared hyperspectral imaging technology (NIR-HSI) (950-1700 nm), a rapid identification method was proposed for Ziziphi Spinosae Semen (Suanzaoren, SZR) and its three kinds of counterfeits, i.e. Ziziphus mauritiana lam (Lizaoren, LZR), Hovenia dulcis Thunb. (Zhijuzi, ZJZ) and Lens culinaris (Bingdou, BD). According to the proportion of 2:1, by randomly dividing the sample set, 480 samples are taken as the training set and 240 samples are taken as the test set. Five preprocessing methods were used to process the extracted raw spectra from region of interest, and the optimal preprocessing method was selected. The full spectral models were established by using the Grey Wolf Optimizer (GWO-SVM), partial least square discrimination analysis (PLS-DA) and soft independent modeling class analog (SIMCA) algorithms. The best classification results of the full spectrum-based PLS-DA, GWO-SVM and SIMCA models were 0.95, 0.99 and 0.97, respectively. Selecting characteristic wavelength by combining spectral data with Competitive adaptive reweighted sampling (CARS) and Successful projects algorithm (SPA) algorithms. The comparison results showed that the recognition rate of SPA-GWO-SVM and SPA-SIMCA were 0.97. The optimal model was SPA-NON-SIMCA. Finally, according to prediction results of the optimal model, the samples were marked with different colours to obtain the visualization map of SZR with different fake products.