Objective  To detect the anthocyanin contents of winegrape skins during ripening stages using hyperspectral imaging technology.

Method  The 900–1 700 nm near-infrared hyperspectral imaging technology and multiple regression methods were used to build prediction models for anthocyanin contents in skins of different winegrape varieties during ripening stage. Totally 75 groups of grape samples belonging to 6 varieties were collected at 4–5 phases of mature stage, and their hyperspectral images were scanned. The spectrum data were enhanced by different preprocessing methods. Dimensionality reduction was then performed by principal component analysis (PCA) and successive projections algorithm (SPA). The anthocyanin contents measured by chemical method were used as reference values, and the prediction models of anthocyanin contents were built using support vector regression (SVR) method.

Result  The SPA-SVR model had the best performance of prediction with the determination coefficient ( R^2_\rm p ) being 0.869 1 and the root mean square error of prediction (RMSEp) being 0.135 9.

Conclusion  It is feasible to use the hyperspectral imaging technology to realize non-destructive and fast detection of the anthocyanin contents in winegrape skins of different varieties during ripening.