Prediction Of The Quality And Harvest Period Of Cabernet Sauvignon Grape Using Near Infrared Spectroscopy

Cabernet Sauvignon grapes are planted extensively.However,in some areas,Cabernet Sauvignon grapes are usually grown in small households,and the quality of the grapes is uneven,which is fundamentally reduced the quality of wine.The quality of wine has led to the loss of economic benefits of wine,making it difficult to gain a foothold in the international market.Many factors affect the quality of Cabernet Sauvignon grapes,among which the content of internal substances and harvest period are direct factors that affect the quality of the grapes.At present,traditional methods are mostly used to determine the quality and harvest perids of Cabernet Sauvignon grapes.These methods have disadvantages such as high cost,slow detection speed,and not suitable for large-scale sample detection.In this paper,Cabernet Sauvignon grapes in different harvest periods were chosed as the research object.The soluble solids,p H,total acid and total phenol content of Cabernet Sauvignon grapes were studied by using near-infrared spectroscopy and chemometric methods.The harvesting period of Cabernet Sauvignon grapes was judged,which provided a certain theoretical basis and technical methods for the development of a portable near-infrared spectrometer dedicated to the quality of Cabernet Sauvignon grapes.At the same time,it is of great significance to the development of Xin Jiang's wine industry.The main research content and results of this paper are as follows:(1)The best pretreatment method corresponding to SSC,p H,total acid and total phenol of Cabernet Sauvignon grapes was studied in the full spectra.Four methods of multivariate scattering(MSC),standard normal transformation(SNV),vector normalization(VN),first derivative+SG smoothing(1D+SG)were used to preprocess the spectral data of Cabernet Sauvignon grapes in different harvest periods.The pre-processed spectra were used to establish PLS and SVR models,and the effects of different pre-processing methods on the PLS model were compared.The results showed that the PLS model with SNV pretreated method had the best performance in predicting p H and total phenol content,with R_pvalues of 0.8712 and 0.8086,RMSEP values of 0.0747 and 2.3642,and RPD values of 2.0368 and 1.6998,respectively.The MSC+PLS model established by MSC algorithm was the most suitable for predicting the total acid content,with the R_P value of 0.8123,RMSEP value of 0.5359 and RPD value of 1.6960.For SSC,the PLS model established by raw spectra had the best prediction performance,with R_P value of 0.8124,RMSEP value of 0.9715 and RPD value of 1.7151.(2)The optimal prediction model of SSC,p H,total acid and total phenol contents in Cabernet Sauvignon grapes was established by using different characteristic wavelength selection methods.Compared and analyzed the effect of CARS,GA,si-PLS,and SPA on the prediction results of the two models of PLS and SVR.According to the results,the best prediction model for SSC was CARS+SVR,with R_P,RMSEP and RPD values of 0.9481,0.4583 and 3.1437,respectively.The best model for predicting p H and total phenols was SNV+CARS+SVR,with R_P values of 0.9440 and 0.8620,RMSEP values of 0.0397 and 2.0283,and RPD values of 2.7459 and 1.9729,respectively.MSC+CARS+SVR was the most suitable model for predicting the total acid content,and the R_p,RMSEP and RPD values were0.8665,0.4043 and 2.0031,respectively.(3)The best pretreatment method corresponding to the harvest period of Cabernet Sauvignon grapes was studied in the full spectra.MSC,SNV,VN and first derivative+SG were applied to preprocess the near-infrared spectra of Cabernet Sauvignon grapes.Preprocessed spectra were used to establish PLS-DA and SVM discriminant models respectively,and the effects of different preprocessing methods were compared and analyzed.According to the predicted accuracy of PLS-DA and SVM models,the best pretreatment method corresponding to the harvest period was selected.The results show that the1D+SG+PLS-DA and 1D+SG+SVM models established by the first derivative+SG method had the best predictive performance,and the discrimination accuracy of the correction set were all 100%.The discriminant accuracy in the prediction set had reached 90%and 100%respectively.(4)The optimal discriminant model of harvest period was studied with using different characteristic wavelength selection methods.CARS,GA,si-PLS,and SPA methods selected wavelength variables of Cabernet Sauvignon grapes,and the effects on PLS-DA and SVM models of three methods were compared and analyzed.Considering the overall performance of the model,the 1D+SG+CARS+SVM model established based on 121 wavelength variables selected by CARS algorithm has the best effect on the determination of harvest periods,with the accuracy of the prediction set reaching 98%and the accuracy of the correction set reaching 100%.