Classification of Tea Characteristics Elements Based on Laser-Induced Breakdown Spectroscopy and Modeling of Fermentation

Technical Summary: Classification of Tea Characteristics Elements Based on Laser-Induced Breakdown Spectroscopy and Modeling of Fermentation

Problem Statement
The tea industry faces significant challenges regarding quality control, including cultivar confusion, batch mixing, and adulteration. Traditional quality assessment relies heavily on sensory evaluation, which is subjective and lacks reproducibility, or physicochemical analysis, which is time-consuming, destructive, and requires complex sample preprocessing. While optical spectroscopy offers a non-destructive alternative, existing Laser-Induced Breakdown Spectroscopy (LIBS) studies have largely focused on coarse discrimination between broad tea categories (e.g., green vs. black tea). There is a notable gap in the literature regarding fine-scale identification of samples from the same tea type but different production batches, the detection of blended samples, and the systematic integration of intrinsic fermentation differences into classification and quantitative modeling.

Methodology
This study proposes a comprehensive framework combining LIBS with advanced chemometric algorithms to address these gaps. The methodology involves three primary stages:

1. Experimental Setup and Sample Preparation:

   • Instrumentation: A Q-switched Nd:YAG laser (1064 nm, 85 mJ, 5 Hz) was used to ablate tea samples. Emission spectra were collected in the 181–835 nm range using a four-channel fiber-coupled spectrometer.
   • Samples: The study utilized 38 tea specimens, including 20 semi-fermented oolong teas and 18 non-fermented green teas. These samples covered diverse cultivars, production batches, pure raw materials, and artificially blended mixtures (1:1 mass ratio). Additionally, post-fermented Pu-erh tea was used for quantitative modeling.
   • Preparation: Samples were ground, sieved (200-mesh), and compressed into standardized pellets (30 mm diameter, 3 mm thickness) under 20 tons of pressure to minimize matrix effects. Each sample underwent 260 laser pulses to ensure signal stability.

2. Data Preprocessing and Feature Selection:

   • Preprocessing: Raw spectra were averaged (every 52 pulses) and processed using Savitzky–Golay (SG) smoothing and background normalization to remove noise and baseline drift while preserving peak characteristics.
   • Feature Selection: The Variable Importance in Projection (VIP) algorithm was employed to screen highly correlated characteristic elements. This reduced dimensionality and isolated spectral peaks sensitive to varietal and fermentation differences.
   • Selected Elements: For oolong tea, nine lines (C, Ca, Na, K, Fe, Al, Mg, Mn, Si) were identified. For green tea, nine lines (C, N, Ca, Na, K, Fe, Al, Mg, Mn) were selected.

3. Modeling Strategy:

   • Classification: A Random Forest (RF) algorithm was constructed using the VIP-selected features to distinguish between major categories (green vs. oolong).
   • Fine-Grained Discrimination: To differentiate between batches and blended samples within the same category, a Pearson correlation coefficient threshold strategy (threshold set at 0.999) was applied. Samples with coefficients above the threshold were classified as identical batches; those below were identified as distinct batches or mixtures.
   • Quantitative Modeling: For Pu-erh tea, the VIP-optimized features were used to build RF regression models to predict core fermentation quality indicators: moisture, water extracts, and tea polyphenols.

Key Results

• Macro Classification: The RF model achieved 100% accuracy in distinguishing between green tea and oolong tea.
• Fine-Grained Identification: The combined approach of VIP feature selection and Pearson correlation analysis successfully achieved 100% accuracy in:
  • Distinguishing distinct production batches of the same tea variety.
  • Identifying pure samples versus artificially blended mixtures.
• Quantitative Prediction: The RF regression models effectively captured the nonlinear correlations between spectral signals and the fermentation quality indicators of Pu-erh tea, demonstrating the ability to quantitatively assess fermentation degree.
• Spectral Correlation: The study confirmed that specific elemental signatures (e.g., Mg, Mn, Al for oolong; Mg, Mn, K, Al for green tea) serve as reliable fingerprints for cultivar discrimination and processing differences, while organic elements (C, N) reflect fermentation-induced degradation.

Significance and Claims
The authors claim that this study provides a reliable technical support system for the intelligent, non-destructive testing of tea. By coupling VIP feature optimization with Random Forest modeling and Pearson correlation analysis, the research successfully mines spectral elemental response patterns corresponding to fermentation differences.

The significance of this work lies in its ability to:

• Overcome the subjective limitations of traditional sensory evaluation.
• Reveal differential response behaviors of various fermentation processes at a microscopic spectral level.
• Provide a unified framework that addresses both category sorting and quality regression, linking classification performance directly to intrinsic fermentation divergence.
• Offer technical methodologies for fermentation process traceability, standardized fermentation grading, mixed sample authentication, and the intelligent control of fermentation quality.

The paper concludes that this approach presents promising prospects for the practical application of standardized optimization in tea fermentation processes.