Abstract: A fast method for the identification of the characteristic components of Chrysanthemum was proposed using three-dimensional excitation emission matrix (3DEEM) spectroscopy coupled with parallel factor analysis (PARAFAC). The 3DEEM spectra of four varieties of Chrysanthemum were obtained and preprocessed to remove interference data such as Raman and Rayleigh scattering as well as outliers, and the spectral characteristics were analyzed. Feature extraction was then performed using PARAFAC, and amino acids and flavonoids were identified as characteristic fluorescent components of Chrysanthemum based on variance interpretation rate and residual analysis. Finally, support vector machine (SVM) and back propagation neural network (BPNN) were employed to analyze the characteristic variables and a fast and non-destructive identification model was developed. The accuracy of the SVM and BPNN models were 100% and 95.93% for the training set, and 94.50% and 89.61% for the test set, respectively. This study demonstrates that 3DEEM-PARAFAC combined with SVM enables both qualitative and quantitative analysis of Chrysanthemum’s characteristic components and rapid identification of Chrysanthemum.

Key words: Chrysanthemum; three-dimensional excitation emission matrix spectroscopy; characteristic component identification; parallel factor analysis; support vector machine; back propagation neural network