RESUMEN
A total of 139 batches of Chrysanthemum samples were randomly divided into calibration set (92 batches) and prediction set (47 batches). The near infrared diffuses reflectance spectra of Chrysanthemum varieties were preprocessed by a first order derivative (D1) and autoscaling, and a modelwas built using partial least squares analysis. In this study, three Chrysanthemum varieties were identified, the accuracy rates in calibration sets of Dabaiju, Huju, and Xiaobaiju are 97.60, 96.65, and 94.70%, respectively; And 95.16, 86.11, and 93.46% accuracy rate in prediction sets was obtained. The research results demonstrate that the qualitative analysis can be conducted by machine learning combined with Near-Infrared Spectroscopy, which provides a new method for rapid and non-invasive identification of Chrysanthemum varieties.