Determination Of Anthocyanin Content Of Winegrapes Skins Using Hyperspectral Imaging Technique

The anthocyanins content of wine grape skin played an important part in color and sensoryquality of wine. However, the customary methods were characterized with time-wasting andexpensive. This work aimed to determine the anthocyanins content in skin based on hyperspectralimaging technology. The grapes of Cabernet Sauvignon (Vitis vinifera L.) produced in Shaanxiprovince were invoked as experimental materials. Hyperspectral images of60grape samples werecollected using near infrared hyperspectral camera (900~1700nm). After then, the anthocyaninscontent of skin was detected by pH-differential method. Grape berries in hyperspectral imageswere extracted as region of interest (ROI) and their average spectrum were calculated. Moreover,different preprocessing methods were used to improve the signal noise ratio (SNR) includingSavitzky-Golay smoothing, normalization and multiplicative scatter correction, et al. Theprediction model for determining anthocyanin content was established using the partial leastsquares regression (PLSR), least squares support vector regression (SVR) and BP neural network(BPNN) with the full-spectrum. Successive projections algorithm was applied to extract effectivewavelengths (EWs), which showed least collinearity and redundancies in the spectral data.Selected effective wavelengths were used as the inputs of multiple linear regression (MLR),partial least squares (PLS) and BP neural network (BPNN). Then the SPA-MLR, SPA-PLS andSPA-BPNN models were developed. The main creative results were achieved as follows:(1) The PLS modeling results showed that the result of MSC is the best compared with theother pretreatments. The BP neural network modeling showed that, compared with the PCA, thePLS had a stronger ability to exact the spectral imformation related to the anthocyanins content.(2) It was demonstrated that the prediction coefficient of determination (P-R2) built by BPNNmodel was0.9102and the root mean square error of prediction (RMSEP) was0.3795. It wasfeasible to measure anthocyanins content in wine grape skins using hyperspectral imaging.(3) Compared with SPA-MLR, SPA-BPNN and SPA-PLS. SPA-PLS was achieved the bestperformance of Rp=0.9000and RMSEP=0.5506. The results indicated that anthocyanins contentin wine grape skins could be measured effectively by utilizing a few fixed bands.