A Machine Learning Method for the Fine-Grained Classification of Green Tea with Geographical Indication Using a...

Area of Science:

• Food Science and Technology
• Analytical Chemistry
• Machine Learning Applications

Background:

• Chinese green tea boasts significant health benefits and diverse categories, including those with geographical indications (GTSGI).
• High-quality GTSGI, often labeled as famous GTSGI (FGTSGI), are distinguished by specific origins but possess subtle differences.
• The fine-grained classification of these highly similar tea categories presents a significant challenge.

Purpose of the Study:

• To develop a novel framework for the fine-grained classification of Chinese green tea sub-categories.
• To accurately classify Maofeng and Maojian green tea categories using electronic nose data.
• To identify famous geographical indications (FGTSGI) within these green tea categories.

Main Methods:

• Proposed a novel Convolutional Neural Network backbone (CNN) combined with a Support Vector Machine classifier (SVM), termed CNN-SVM.
• Utilized electronic nose data, constructing a multi-channel input matrix for the CNN backbone to extract deep features.
• Employed an SVM classifier for enhanced discrimination, particularly effective for small sample sizes, and compared performance against other models.

Main Results:

• The proposed CNN-SVM framework achieved superior performance in classifying GTSGI and identifying FGTSGI compared to other machine learning models.
• Demonstrated high accuracy and strong robustness in distinguishing between multiple, highly similar Chinese green tea sub-categories.
• The framework effectively leveraged deep feature extraction from sensor signals for precise classification.

Conclusions:

• The CNN-SVM framework offers a highly effective solution for the fine-grained classification of Chinese green tea categories.
• This approach shows significant potential for accurately identifying and authenticating high-value teas, including FGTSGI.
• The study highlights the capability of electronic nose technology combined with advanced machine learning for tea quality assessment.