Optimization Of Hyperspectral Algorithm For Rapid Detection Of Resveratrol Content In Wine

The traditional detection methods of resveratrol in red wine are time-consuming,cumbersome,detrimental and polluting the environment.They can not meet the development requirements of rapid,non-destructive and real-time of red wine detection.In this study,a circulating fluidized bed enrichment-hyperspectral imaging detection device was designed.The red wine(Cabernet Gernischt wine)was used as the object of study,the resveratrol was enriched by the fluidized bed device,the image and spectral information of samples was collected by hyperspectral imaging.Based on chemometrics method,image processing algorithm and data mining technology,the hyperspectral detection model of resveratrol content in red wine was established and optimized algorithm,which could provide theoretical support for trace element detection in wine using hyperspectral technology.The main research contents and achievements were as follows:1)Elimination of abnormal wine samples.The Hotelling T2 detection method of abnormal samples was used to eliminate abnormal samples,and the PLSR model was built,the Rcv,2 was 0.7312,the RMSECV was 0.0562;the Rp2 was 0.7050,the RMSEP was 0.7050.The precision of the original data was improved by the elimination of interference data.2)Red wine sample-set Partitioning.The results of red wine sample-set Partitioning on KS,SPXY and RS methods were compared.Among them,the maximum precision of the model was KS method,the prediction set RP2 was improved by 0.0637 and the RMSEP was reduced by 0.006.The reasonable partitioning method of the sample set could further improve the prediction accuracy,and the optimal partitioning method of the red wine sample set was KS method.3)Spectral pretreatment of resveratrol content.The result of the PLSR model showed that the SNV pretreatment method was the best among pretreatment methods of MSC,SG and SNV when comparing the parameters of the red wine.the RP2 was 0.7824 and the RMSEP was 0.0502.Compared with the prediction model of original data,the RP2was increased 0.0137,the RMSEP was reduced 0.0029.Pretreatment methods could eliminate irrelevant information and improve the model stability.4)Establishment and evaluation of hyperspectral quantitative analysis model of resveratrol content.Comparing the effects of MLR,PCR,PLSR and SVMR,the prediction effect of PLSR-SNV model was optimized.The prediction set RP2 was 0.7824 and RMSEP was 0.0502.It is feasible to detect the resveratrol content in red wine by near hyperspectral imaging technology.5)Comparison of Hyperspectral Characteristic wavelength of resveratrol content.The effects of SPA,UVE and CARS which chose hyperspectral characteristic wavelength of resveratrol content were compared.SPA was selected for the spectral band selection,the prediction set RP2 was 0.7167,the RMSEP was 0.0578.and among these three characteristic wavelength extraction methods,SPA was superior to UVE and CARS.In the whole band,the RP2 was 0.7824 and the RMSEP was 0.0502,and the results of the full band algorithm were optimized.