Studies On The Digital Quality Controlling Of Tea Products Based On Multi-spectral Technology

In recent years,the tea market has frequently appeared fake and inferior phenomenon.Sugar and syrup were illegally added in the process of tea production,stored tea was sold as fresh tea,and tea grades were randomly labeled.Such unscrupulous competition,not only affected the market order,but also harm the tea brand and damage the rights and interests of consumers.Morever,the additives used in processing are also threating food safety.Given the traditional sensory evaluation and chemical indicators detection methods can not meet the needs of online detection.Based on the digital quality control technology,we attempt to establish a scientific,simple and comprehensive evaluation method for tea quality and safety.Three kinds of techniques,including near infrared spectroscopy,hyperspectral imaging and olfactory sensors,were used to obtain the multi-directional information of tea.Combined with multivariate data analysis method,the discrimination of illegal addition in tea,the identification of tea storage period and the evaluation of tea grade were studied.The main research results include the following aspects:1.The sample preparation conditions of near infrared spectroscopy were optimized,and the optimal sample preparation conditions were obtained:particle size was 40-60 mesh,pressure condition was 40 MPa,and sample thickness was 4 mm.Based on statistical analysis of the external validation results of NIR model,the best modeling spectral range and pretreatment method were optimized.Roasted green tea doped with sugar samples modeling optimal spectral range was 7502-6098.1cm-1 and 5450-4597.7cm-1,the optimal pretreatment method was subtracting a straight line method;roasted green tea doped with glucose syrup samples modeling optimal spectral range was 9403.6-8450.9 cm-1 and 6101.9-4597.7 cm-1,the optimal pretreatment method was min-max normalization method.PLS was utilized to establish the quantitative model.For roasted green tea doped with sugar samples,R2=99.76,RMSECV=0.313 in calibration set,R2=99.56,RMSEP=0.432 in prediction set.For roasted green tea doped with glucose syrup samples,R2=99.6,RMSECV=0.408 in calibration set,R2=99.79,RMSEP=0.297 in prediction set.PCA-Euclidean distance was utilized to establish the qualitative model.For roasted green tea doped with sugar samples,the discrimination rate was 96%.For roasted green tea doped with glucose syrup samples,the discrimination rate was 100%,which achieve accurate identification of tea doping.2.On the basis of the cost and convenience of the instrument,an integrated spectrometer with a diffraction grating,a digital micromirror device as programmable wavelength filter,and InGaAs detector was bought from Texas Instruments.In the experiment,the model and software platform were built,and a portable NIR spectroscopic system was developed.The software part of data processing is based on GA-PLS algorithm,which can realize accurate analysis of tea doping.3.Based on the analysis of artificial aging tea samples,the stored tea samples and the fresh-keeping stored tea samples,the model of tea storage period was established by near infrared spectroscopy(NIR).PCA-KNN,PCA-LDA and SVM algorithms were utilized and compared.The discriminant rate of PCA-LDA model with artificial aging roasted green was 100%,and the discriminant rate of SVM model with artificial aging Huangshan Maofeng was 97.83%.Besides,the discriminant rate of SVM model with stored Lu'an Guapian was 92%,and the discriminant rate of SVM with stored Huangshan Maofeng was 92%.The modeling results can basically meet the discriminant needs.However,the discriminant rate of PCA-KNN model with fresh-keeping stored Lu'an Guapian was 82%,and the discriminant rate of PCA-LDA model with fresh-keeping stored Huangshan Maofeng was 80%,which indicated the single spectral information can not realize the accurate identification of the more complex fresh-keeping storage tea.4.Based on the analysis of artificial aging tea samples,stored tea samples and fresh-keeping stored tea samples,the model of tea storage period was established by HSI.With principal component analysis,five characteristic wavelengths were selected:670.74,720.08,836.14,886.09,and 936.05 nm.In storage time discrimination of stored tea samples,the texture data modeling effect was better than the spectral data.In addition,the modeling effect of data fusion model was better than single eigenvalue.It showed that the SVM algorithm has obvious superiority in model establishment.In storage time discrimination data fusion model,the discriminant rate of stored Huangshan Maofeng was 98%,and the discriminant rate of stored Lu'an Guapian was 96%.Besides,the discriminant rate of fresh-keeping stored Huangshan Maofeng and Lu'an Guapian were both 100%.In addition,the corresponding relationship between the stored tea sample,the artificial aging tea sample and the fresh-keeping stored tea sample can be found by using GA-PLS.5.Based on hyperspectral image technology,the SVM model of tea grading was established.The effects of different scanning cameras(scanning bands)and different sample morphologies were compared.The results show that based on VNIR/NIR band,the discriminant rate of tea powder model was 98.7%in the prediction set.However,the tea powder sample basically loses the image characteristic.Based on VNIR/NIR band,the discriminant rate sample of tea model was 8 80%in the prediction set,which was the best model based on the texture data.In addition,based on spectral data in NIR band,the discriminant rate was 91.25%in the prediction set.The tea classification requirement basically satisfied.Furthermore,five characteristic wavelengths of the VNIR/NIR band were optimized:670.74,769.70,825.05,880.54 and 936.05 nm,five characteristic wavelengths of NIR band were optimized:1102.25,1232.78,1314.38,1485.79 and 1567.44 nm.6.An innovative and inexpensive colorimetric sensor array-based artificial olfactory system was developed for intelligent evaluation of green tea's grade.The colorimetric sensors array was man-made using printing 12 chemically responsive dyes(9 porphyrins,metalloporphyrins and 3 pH indicators)on silica-gel flat plate.The gel flat plate were exposed with VOCs and the colour changes in each sample were obtained by distinguishing between the images of sensor array before and after contact with tea sample.The values of colour composition changes were extracted from the dyes colour sections.By using image filtering and threshold segmentation,36 RGB feature variables were extracted from the difference image.In the experiment,the adjacent grade tea samples had the tendency of clustering distribution in the PCA scatter map,which indicated the similarity in tea aroma components of adjacent grades.As a supervised algorithm BP-ANN was superior to PCA algorithm,and the discriminant rate can reach 85%and 86%in calibration and prediction set,respectively.All misclassification samples in the model were wrongly divided into adjacent grades,which were similar to PCA results.7.Based on hyperspectral image technique,the texture feature was extracted at the characteristic wavelength(670.74,769.70,825.05,880.54,936.05 nm)in VNIR/NIR band,and the spectral feature was extracted at the characteristic wavelength(1102.25,1232.78,1314.38,1485.79,1567.44 nm)in NIR band.Based on olfactory sansor,the RGB feature was extracted from difference images.The characteristic layer is fused with different eigenvalues.Compared with PCA-KNN and Pca-lda,SVM shows a good ability to deal with complex data,and shows obvious advantages in fusing data of different sensors.The discriminant rate of SVM model based on data fusion and predictive se(?)odel is 92%,which is better than the single information model based on spectrum,texture or RGB characteristic value.Three different types of sensor information complement each other,effectively improve the accuracy of discrimination,to achieve a comprehensive evaluation of tea grade.Different features fused at the feture layer.PCA-KNN,PCA-LDA and SVM algorithms were utilized and compared.Among them,SVM showd good ability to handle complex data in the fusion of different sensors data.The discrimination rate of SVM model based on fusion data was 92%in calibration and prediction set,respectively.It indicated the fusion model was superior to single feture model based on spectral,texture or RGB eigenvalues.Overall,we conclude that combined three-feature information derived from two sensors can be used to realize the accurate identification of tea grades.8.The aroma components in tea sample were quantitative detected by GC-MS.Based on the response of GC-O and olfactory sensor array,7 aroma components were selected and then analyzed by ANOVA single factor analysis.The results of variance analysis showed:cis-3-hexane,benzyl alcohol,aromatic alcohol,benzene ethanol,methyl salicylate,decyl aldehyde,?-ionone have significant difference between different kinds of tea.The relationship between gas sensitive materials and aroma components was validated by using aroma substance monomer and Pearson correlation analysis.The correlation analysis showed that the content of aromatic alcohol was significantly correlated with the response value of multiple gas sensors,and the content of benzene ethanol and decyl aldehyde was significantly correlated with the response value of one gas sensor.