Abstract:

This work aimed to determine the anthocyanin content in grape skins based on hyperspectral imaging
technology in combination with successive projections algorithm (SPA). Cabernet Sauvignon (Vitis vinifera L.) grape berries
from Shaanxi province were used as experimental materials. Hyperspectral images of 60 groups of grape samples were
collected by near infrared hyperspectral camera and the anthocyanin contents in these samples were detected. Multiplicative
scatter correction was used to improve the signal-to-noise ratio (SNR). Moreover, SPA was applied for the extraction of
effective wavelengths (EWs), which showed least collinearity and redundancies in the spectral data. The selected effective
wavelengths were used as the inputs of multiple linear regression (MLR), partial least squares (PLS) and BP neural network
(BPNN). Then SPA-MLR, SPA-PLS and SPA-BPNN models were developed and compared with full-spectrum-PLS
model. It was shown that SPA-MLR, SPA-PLS and SPA-BPNN models were better than full-spectrum-PLS model. The
best performance was achieved by SPA-PLS model with Rp of 0.900 0 and RMSEP of 0.550 6. These results indicate that
anthocyanin contents in grape skins could be measured effectively by using near infrared hyperspectral imaging.

Key words: winegrape, anthocyanin, hyperspectral image, successive projections algorithm (SPA), partial least squares (PLS)