Abstract: In this experiment, fluidized-bed enrichment of trace resveratrol in red wine was carried out by macroporous resin adsorption and hyperspectral images of the resin samples were acquired. The prediction models established using various spectral pretreatments were compared for obtaining the optimal algorithm. The results showed that the partial least squares regression (PLSR) model established by removing abnormal samples using Hotelling T2 test method, dividing the sample sets using the KS algorithm, and pretreating the spectral data using standard normal variate (SNV) method exhibited the best prediction performance with correlation coefficient of correction (Rc2), root mean square error of calibration (RMSEC), correlation coefficient of prediction (Rp2) and root mean square error of prediction (RMSEP) of 0.813 8, 0.047 7, 0.782 4, and 0.050 2, respectively. Hyperspectral imaging can provide a new method for detecting trace components.

Key words: hyperspectral technology, resveratrol, fluidized bed, enrichment, algorithm optimization