Abstract:The hyperspectral data （900-1 750 nm） of non-polarization and at four polarization angles （0°，45°，90°，and 135°） were collected for winter jujube from South Xinjiang and red grapes （used for validation）.The original spectra were processed by envelope removal. The competitive adaptive reweighted sampling algorithm was used for the dimension reduction of data，and the most effective wavelengths were selected. A partial least squares regression（PLSR） model was established with the reflectance data from one non-polarized and four polarized directions to predict the moisture content and soluble solid content（SSC） in winter jujube from South Xinjiang. Compared with non-polarized hyperspectral modeling，the models with data from spectral data with the polarization angles of 90° and 135° showed improved performance in predicting the moisture content and SSC of winter jujube，achieving correlation coefficients of 0.958 8 and 0.924 3，respectively，and remaining prediction deviations greater than 2. Similar results were obtained in the modeling for red grapes. The findings demonstrated that polarized hyperspectral imaging outperformed non-polarized hyperspectral imaging in modeling. The modeling with the data from hyperspectral imaging at the polarization angles of 90° and 135° had the highest accuracy in predicting the moisture content and SSC of winter jujube.