Study On Grapes’ Internal Quality Detection Base On Hyperspectral Imaging Technology

Xinjiang grapes are delicious and have rich nutritional value, they’re also popular for the majority of consumers.Meanwile, the quality of the grapes determines their own commercial value and market competitiveness. The traditional detection methods are not conducive to the commercialization of grape after postharvset, because they used for grape internal quality are time-consuming and destructive. In this paper,specialty grape of Xinjiang(red grape) was used as the research object, hyperspectral imaging technology, chemometrics statistical and mathematical knowledge were used to carry out internal quality detection of grapes(soluble solids content and total acidity), which provide a new idea for the grape industry with in real-time, on-line, non-destructive detection. The main contents and results of this study are listed as follows:(1)Abnormal samples were identified and removed to make soluble solids content and total acidity models of grapes representative and stability in the participated modeling. Dixon detection, Leverage and student residual were used to judge comprehensively, suspected abnormal samples were searched and recovered one by one analysis(twice-detection). Ultimately abnormal samples of soluble solids content modeling and total acidity modeling of grapes were determined and removed.(2)The sample used for modeling were divided according to KS, SPXY and CG method, comparative analysis the statistics results of calibration set and prediction set and PLS model performance after sample partition by different methods, it showed that SPXY was the best method to divide the soluble solids content calibration set and CG was the best method to divide the total acidity calibration set in grape.(3)Through different pretreatment methods to preprocess the original spectrum of soluble solids content and total acidity in grape, respectively, different prediction models were developed and the performance of models were compared based on PLS, SMLR and PCR. The results of extracting the characteristic wavelength of soluble solids content and total acidity were compared by GA and SPA wavelength selection method, and the model performances of extracting wavelengths, such as the full spectrum, GA and SPA of grape soluble solids content, total acidity value were compared, respectively.1)The result showed that the predictive ability of PLS method was the best among the soluble solids content models after the pretreatmnet of mean centering. When the factor was 14 of PLS model, model performance parameters RC was 0.9616, RP was 0.9606, RMSEC was 0.368, RMSEP was 0.367, RPD was3.5980; Then GA was determined as the best characteristic wavelength selection method combine with PLS to develop SSC model, when the factor was 13 of PLS model, model performance parameters RCwas0.9654, RP was 0.9610, RMSEC was 0.352, RMSEP was 0.298, RPD was 3.6254, the performance of built PLS model was better than PLS model of full spectrum. 2)The result showed that the predictive ability of PLS method was the best among the total acidity models after the pretreatmnet of normalization. When the factor was 8 of PLS model, model performance parameters RC was 0.8737, RP was 0.8871, RMSEC was0.0126, RMSEP was 0.0133, RPD was 2.1665; Then GA was determined as the best characteristic wavelength selection method combine with PLS to develop TA model, when the factor was 7 of PLS model,model performance parameters RCwas 0.8765, RPwas 0.8628, RMSEC was 0.0126, RMSEP was 0.0123,RPD was 1.9778, the predictive ability of built PLS model was slightly less than full-spectrum PLS model,but the stability has improved.