Identification of Tea Storage Times by Linear Discrimination Analysis and Back-Propagation Neural Network Techniques Based on the Eigenvalues of Principal Components Analysis of E-Nose Sensor Signals

Huichun Yu, Yongwei Wang, Jun WangSensors 2009 · DOI

TL;DR

This paper explores using an electronic nose (E-nose) combined with Principal Component Analysis (PCA), Linear Discriminant Analysis (LDA), and Back-Propagation Neural Networks (BPNN) to identify the storage time of green tea based on its aroma. The study demonstrates that analyzing the E-nose signals from tea leaves, after dimensionality reduction with PCA, provides the most accurate discrimination and prediction of storage duration.

What Problem Does This Paper Address?

The quality of green tea, like many food products, degrades over time due to oxidation and loss of beneficial components, which directly impacts its aroma. Traditionally, tea quality is assessed by human sensory panels, a method prone to subjectivity and variability. This paper addresses the need for an objective, reliable, and non-destructive method to assess tea quality by quantifying changes in its aroma over extended storage periods. Specifically, it aims to determine how long green tea has been stored by analyzing its volatile organic compounds (VOCs) using an E-nose.

Methods

The researchers employed a multi-stage approach:

  1. Electronic Nose (E-nose) Data Acquisition: A portable E-nose (PEN2) with 10 metal oxide sensors was used to capture the aroma profiles of tea samples. Headspace gas from the tea samples was pumped into the sensor chamber, and the conductance change of each sensor was recorded over 60 seconds.
  2. Feature Extraction: The raw sensor data was processed to extract eight different feature values per sensor, capturing various aspects of the response curve (e.g., initial response, peak, stable stage, integral). This resulted in an initial feature vector of 80 dimensions (10 sensors x 8 features).
  3. Dimensionality Reduction with Principal Component Analysis (PCA): To manage the high dimensionality of the E-nose data and reduce noise, PCA was applied. The top five principal components, which captured over 80% of the variance, were selected as the optimized feature vector.
  4. Discrimination Analysis with Linear Discriminant Analysis (LDA): LDA was used to classify the tea samples based on their storage times using the PCA-reduced feature vectors. LDA aims to maximize the variance between different storage time groups while minimizing variance within each group.
  5. Prediction with Back-Propagation Neural Network (BPNN): A BPNN was employed to predict the actual storage time of the tea samples. The five principal components served as inputs, and the output was the predicted storage duration.

Datasets

The study utilized Longjing green tea (AAA grade) produced on July 1, 2005. Tea samples were stored under cold storage (4°C) and analyzed at intervals of 0, 60, 120, 180, and 240 days. The experiments involved analyzing three types of samples:

  • Tea Leaves: 45 packages of tea leaves were used for each detection.
  • Tea Beverages: Prepared by brewing 5g of tea leaves with 250mL of boiled water (1:50 ratio) and filtered after 5 minutes.
  • Tea Residues: The solid material remaining after brewing the tea.

For both LDA and BPNN, the samples were randomly divided into a training set (150 samples) and a test set (75 samples).

Key Insights and Ideas

  • Aroma as a Reliable Indicator of Tea Age: The study strongly supports the idea that changes in tea aroma, detectable by an E-nose, are a significant indicator of its storage time and, by extension, its quality.
  • PCA is Crucial for E-nose Data Preprocessing: Reducing the high dimensionality of E-nose data using PCA is essential for improving the efficiency and effectiveness of subsequent classification and prediction tasks. Selecting the top five principal components effectively balances information preservation and dimensionality reduction.
  • Tea Leaves Offer Superior Discrimination: Analyzing the E-nose signals from dry tea leaves yielded significantly better results for both LDA discrimination and BPNN prediction compared to analyzing tea beverages or residues. This suggests that the volatile compounds released directly from the leaves are more sensitive indicators of aging.
  • Synergy of Techniques: The combination of E-nose technology with PCA, LDA, and BPNN provides a robust framework for objective quality assessment of perishable food products.

Key Results

  • LDA Discrimination: LDA successfully discriminated between different storage times for tea leaves, with distinct clusters observed for each age group. The discrimination for tea beverages and residues was less clear, with some overlap between samples.
  • BPNN Prediction Accuracy: The BPNN model demonstrated the best prediction accuracy when using the PCA-derived features from tea leaves. The mean prediction errors for tea leaves were:
  • 0 days: 9 days
    • 60 days: 2.73 days
    • 120 days: 3.93 days
    • 180 days: 6.33 days
    • 240 days: 6.8 days These errors were notably smaller than those obtained for tea beverages and tea residues, which had mean errors ranging from 5.8 to 14.2 days.

Limitations and Future Work

  • Limitations:
    • The study focused on Longjing green tea; the findings might not be directly generalizable to all tea types or grades without further validation.
  • While the BPNN showed good prediction, the mean error for the 0-day sample was relatively high (9 days), suggesting room for improvement in predicting very fresh samples.
  • The paper mentions that the choice of five principal components was "justified by the results (Table 1)," implying a degree of empirical selection rather than a strictly theoretical derivation.
  • Future Work: The authors suggest that this approach could be extended to assess other quality parameters of tea and potentially other food products. Further research could explore different E-nose sensor arrays, advanced pattern recognition algorithms, and investigate the impact of various storage conditions (beyond cold storage) on tea aroma.

Summary updated

Abstract

An electronic nose (E-nose) was employed to detect the aroma of green tea after different storage times. Longjing green tea dry leaves, beverages and residues were detected with an E-nose, respectively. In order to decrease the data dimensionality and optimize the feature vector, the E-nose sensor response data were analyzed by principal components analysis (PCA) and the five main principal components values were extracted as the input for the discrimination analysis. The storage time (0, 60, 120, 180 and 240 days) was better discriminated by linear discrimination analysis (LDA) and was predicted by the back-propagation neural network (BPNN) method. The results showed that the discrimination and testing results based on the tea leaves were better than those based on tea beverages and tea residues. The mean errors of the tea leaf data were 9, 2.73, 3.93, 6.33 and 6.8 days, respectively.
BibTeX
@inproceedings{yu2009identification,
  author = {Huichun Yu and Yongwei Wang and Jun Wang},
  title = {{Identification of Tea Storage Times by Linear Discrimination Analysis and Back-Propagation Neural Network Techniques Based on the Eigenvalues of Principal Components Analysis of E-Nose Sensor Signals}},
  booktitle = {Sensors},
  year = {2009},
  doi = {10.3390/s91008073},
}
Figure