Elsevier

Journal of Food Engineering

Volume 263, December 2019, Pages 165-172
Journal of Food Engineering

Application of machine learning algorithms in quality assurance of fermentation process of black tea-- based on electrical properties

https://doi.org/10.1016/j.jfoodeng.2019.06.009Get rights and content

Highlights

  • Electrical properties used to optimize the fermentation of black tea.

  • Hierarchical clustering provided an objective classification for fermented samples.

  • High correlations have been found between electrical properties and catechins.

  • Random forest can effectively distinguish the degree of fermented samples.

Abstract

Fermentation process directly determines the product quality of black tea. This work aimed to develop a rapid method for detecting the degree of fermentation of black tea based on electrical properties of tea leaves. An LCR meter employed to identify 11 electrical parameters of tea leaves during the fermentation process, and the content of catechins and tea pigments in tea leaves were measured by using HPLC and UV-Vis spectrometer, respectively. Principal component analysis and hierarchical clustering analysis applied to divide samples into different groups in the degree of fermentation. Correlation analysis used to characterize the responding strength of electrical parameters on the variation of catechins and pigments. Finally, multilayer perceptron, random forest, and support vector machine algorithm used to build discrimination models of fermentation degree, and the average accuracy rate on the testing set reached to 88.90%, 100%, and 76.92%, respectively.

Introduction

Black tea is one of the most widely consumed beverages in the world (Bhattacharyya et al., 2007a, Bhattacharyya et al., 2007b; Li et al., 2013; Yassin et al., 2015). In general, the production process includes five steps, withering, rolling, fermentation, drying and sorting (Lee et al., 2016; Samanta et al., 2015). The fermentation process is the most critical step due to the powerful effect on the flavour and quality of black tea. While over-fermenting of black tea generates winey or sour flavour, under-fermenting may result in green infusion, raw or thin mouthfeel (Bhattacharyya et al., 2007a, Bhattacharyya et al., 2007b). Tea catechins were oxidized during the fermentation process under the catalysis of polyphenol oxidase and peroxidase and formed two significant groups of pigments (Kusano et al., 2015; Vargas and Vecchietti, 2016), the orange-red theaflavins (TFs), and the reddish-brown thearubigins (TRs) (Stodt et al., 2014). TFs present briskness, strength, and brightness to tea infusion, TRs contribute red colour to tea leaves and infusion (Ghosh et al., 2012). Both of TFs and TRs are considered as the quality index parameters in black tea (Kumar et al., 2011; Ghosh et al., 2015). Theabrownins (TBs) is another kind of pigments that present brown or reddish-brown, and formed (oxidation, polymerization, coupling) from polyphenols, TFs and TRs (Gong et al., 2012). Traditionally, TBs produced negative impacts in tea infusion such as becoming dark and turbidity, and TBs accumulation closely associated with over-fermenting. Hence, monitoring the conversion from catechins to pigments (TFs, TRs, and TBs) which are conducive to master the veracious progress of fermentation in tea leaves.

Electrical properties have been widely applied in the food industry for many years, such as moisture measurement in grains (Ahmed et al., 2007; Sacilik et al., 2007), fruit maturity determination (Sosa-Moralesa et al., 2009; Guo et al., 2007), as well as meat and seafood freshness examination (Nelson, 2008). In dairy sector, electrical properties have been used in milk freshness assessment (Mabrook and Petty, 2002), fat content measurement (Żywica et al., 2012), adulteration identification (Banach et al., 2012), and the evaluation of the rind percentage and ripeness of cheese (Cevoli et al., 2015). These research confirmed the fact that food composition included moisture, carbohydrate, ash, and protein can be effectively presented by electrical signals (Jha et al., 2011; Ma et al., 2016). Therefore, it expected that catechins oxidation and pigments formation during the fermentation process of tea leaves could be caught by measurement of electrical properties.

With the fermentation time elapsed, tea leaves from green change to coppery brown, and the odour from grassy converts into floral or fruity smell (Bhattacharyya et al., 2007a, Bhattacharyya et al., 2007b). Artisans usually according to the changes in tea colour and smell to assess the fermentation degree, but it was very subjectivity since it could be affected such as experience, mental state, environment. In order to get more objective judgment, machine vision technique (Borah and Bhuyan, 2005; Dong et al., 2018), electronic nose technique (Bhattacharyya et al., 2007a, Bhattacharyya et al., 2007b; Sharma et al., 2015), electronic tongue technique (Ghosh et al., 2015) were employed to monitor fermentation process. However, these new techniques used in the tea industry are finite due to the exorbitant cost and complex operation.

The present research suggested and discussed a new approach to evaluate the fermentation degree of black tea based on the electrical properties of tea leaves. An LCR meter performed to measure the electrical properties of tea leaves. The change of catechins and pigments used to cluster analysis by principal component analysis (PCA) and hierarchical clustering analysis (HCA). PCA provided a corresponding relationship of the fermentation process between fermentation time and degree, and HCA produced an objective grouping in fermentation degree of samples. The next was creating discrimination models of fermentation degree by multilayer perceptron neural network (MLP), random forest algorithm (RF) and support vector machine algorithm (SVM), used electrical properties as independent variables, and the HCA clustering results as dependent variables. The accuracy of models used to compare the performances of different algorithm models.

Section snippets

Chemicals and reagents

Individual catechin (C, ≥ 99%), epicatechin (EC, ≥ 98%), epicatechin gallate (ECG, ≥ 98%), epigallocatechin (EGC, ≥ 95%), epigallocatechin gallate (EGCG, ≥ 95%) were purchased from Sigma-Aldrich (St. Louis, MO, USA). The other chemical reagents used were of HPLC grade (Sinopharm, Beijing, China). The Milli-Q water was prepared by an EASYPure UV Ultra Pure Water System (Barnstead International, Dubuque, IA, USA).

Sample preparation

Clonal tea leaves of the Yingshuang variety (one bud with two leaves) harvested from

Data clustering with PCA and HCA

PCA is a dimension reduction process by transforms the original set of variables to a new set of uncorrelated variables, and the new variables are called principal components (PCs) (Yi et al., 2015). PCA could present a better visualization of differences among the various samples by plotting the PCA scores. In this work, all samples with different content of catechins and pigments were evaluated by PCA to check possible sample grouping of fermentation degree. The first principal component

Conclusion

Based on the changes in catechins and pigments, fermented tea samples can be efficiently grouped in chronological order and fermentation degree by PCA and HCA. The high correlation between chemical components (catechins and pigments) and electrical properties was revealed during the fermentation process of black tea. RF presented satisfactory prediction accuracy as a pattern recognition method based on “ensemble learning” strategy, it a good deal with the classification problem in unbalanced

Declarations of interest

The authors declared that they have no conflicts of interest to this work. We declare that we do not have any commercial or associative interest that represents a conflict of interest in connection with the work submitted.

Acknowledgements

The work was supported by Earmarked Fund for China Agriculture Research System [CARS-19]; and the Primary Research and Development Plan of Zhejiang Province [2015C02001].

References (39)

Cited by (0)

Hongkai Zhu and Fei Liu contributed equally to this work.

View full text