Study On The Fermentation Quality Of Black Tea Based On Image And Odor Fusion Monitoring

Black tea is one of the most popular nonalcoholic beverages.The processing process includes withering,rolling,fermentation,and drying.Fermentation directly determines the quality of black tea.Enzymatic oxidation reactions such as tea polyphenols and the degradation of chlorophyll cause changes in leaf color and odor during the fermentation process of black tea.In traditional processing procedures,workers judge the degree of fermentation of black tea by observing the color and odor of black tea,which is highly subjective.In this study,a rapid non-destructive testing technique based on machine vision and e-nose was used to qualitatively analyze the fermentation of black tea.The effects of fermentation time on the nutrients and volatile substances in black tea were studied;Combining machine learning and deep learning,a discrimination model for the degree of tea fermentation was established,and the discrimination effects of single sensor and multiple sensors on the degree of black tea fermentation were discussed;The effects of multi sensor discrimination models with different data fusion strategies were compared,and the objective and scientific discrimination of black tea fermentation was achieved.The main research contents include:(1)Firstly,the effects of different fermentation times(0-7 h)on non-volatile and volatile substances in black tea were investigated.The key substances such as tea polyphenols,catechins,and tea pigments determine the taste of black tea and the quality of tea soup.This study measured the nutrient content at different fermentation time points and explored its changes with the extension of fermentation time;GC-MS was used to detect the changes in volatile compounds during fermentation,a total of 117 common aroma components were detected,including 18 aldehydes,24 alcohols,26 esters,16 ketones,15 alkenes,5 acids,and16 other compounds.The changes in its content can be summarized into three situations:increase or first increase and then decrease,with the peak value mostly at 5h(mostly related to the unique flavor of black tea),decrease(mostly related to green grass gas),and no significant changes.According to the changes in the content of nonvolatile substances,volatile substances and sensory evaluation score with the increase of fermentation time,the optimal fermentation time was determined,and tea fermentation was divided into three stages: insufficient fermentation(0-4 h),moderate fermentation(4.5-5 h),and excessive fermentation(5.5-7 h).(2)The tea color feature values(R,G,B,H,S,V,L,a*,and b*)of different fermentation degrees were extracted by machine vision.As the fermentation time prolonged,the R,G,S,V,L,and b * values decreased.Except for the S value,other feature values mainly changed between insufficient fermentation and moderate fermentation.The a * value rised first and then falled,reaching the highest value during moderate fermentation,and the B and H values did not change significantly;The e-nose was used to collect odor information during the fermentation process of black tea.The sensors with a response value greater than 2 include R2,R6,R7,R8,and R9.During the fermentation process of black tea,the odor response value of tea leaves generally shows a downward trend,with a gradual change in 3-5 hours,and then continued to decline.This may be related to the concentration change of characteristic volatiles representing the aroma of grass,flowers,and fruits.Using image information and odor information as independent variables,and the three stages of tea fermentation as dependent variables,the discriminant models for the degree of black tea fermentation were established by combining random forest(RF),K-nearest neighbor(KNN),and support vector machine(SVM).In addition,a single factor analysis and partial least squares discriminant analysis(PLS-DA)were performed on GC-MS data,and a total of 51 volatile compounds that contributed significantly to the changes in aroma components were selected,including geraniol,linalool,and so on,with a P value of < 0.05 and a VIP value of ≥ 1.(3)Three different data fusion strategies,namely,data level,feature level,and decision level,were used to establish multisensor based black tea fermentation discrimination models,and the effectiveness of the models were compared with that of a single sensor.In feature level fusion,principal component analysis(PCA)and Pearson correlation analysis were used to select appropriate eigenvalues for modeling;Using the powerful data mining and learning capabilities of neural networks combined with one-hot coding to replace expert systems,the basic probability assignment function(BPA)for different samples was obtained,achieving decision level fusion.The results showed that the data fusion strategy can combine different sensor information to obtain more comprehensive feature data,and the model effects were better than that of a single sensor.