Using near-infrared hyperspectral imaging with multiple decision tree methods to delineate black tea quality

Area of Science:

• Agricultural Science
• Food Science
• Spectroscopy

Background:

• Current black tea quality evaluation relies on subjective and empirical human panel tests.
• Developing an objective, automated method for tea quality assessment is highly significant.

Purpose of the Study:

• To develop an objective analytical approach for assessing black tea quality and rank using near-infrared hyperspectral imaging (HSI).
• To compare the effectiveness of different decision tree algorithms for tea quality classification.

Main Methods:

• Near-infrared hyperspectral imaging (HSI) was employed as the primary data acquisition tool.
• Data fusion integrated texture features (Gray-Level Co-occurrence Matrix - GLCM) with spectral features.
• Three supervised decision tree algorithms (fine, medium, coarse tree) were utilized for modeling.

Main Results:

• Data fusion significantly enhanced model performance compared to using single feature types.
• The fine tree model, utilizing fused data, achieved the highest predictive accuracy.
• The best model demonstrated a correct classification rate (CCR) of 93.13% for black tea quality evaluation.

Conclusions:

• Hyperspectral imaging (HSI) combined with intelligent algorithms offers a rapid and effective strategy for objective black tea quality assessment.
• This approach successfully identifies and ranks black tea quality, overcoming traditional subjective methods.