Intelligent prediction of black tea withering degree using grad-CAM-CNN assisted VNIR-colorimetric sensor array

Area of Science:

• Food Science and Technology
• Analytical Chemistry
• Artificial Intelligence

Background:

• Black tea aroma quality is critically dependent on accurate withering degree prediction.
• Current methods for assessing tea withering may lack efficiency and real-time applicability.
• Volatile organic compounds (VOCs) play a key role in tea aroma and are released during withering.

Purpose of the Study:

• To develop an intelligent platform for predicting black tea withering degrees.
• To integrate colorimetric sensor arrays (CSAs) with visible near-infrared (VNIR) spectroscopy.
• To utilize interpretable deep learning models for enhanced accuracy and process control.

Main Methods:

• Selected eight dyes from sixteen candidates based on correlation analysis (R² > 0.36) for CSA development.
• Designed a CSA by drop-casting selected dyes onto filter paper to capture VOCs.
• Employed a convolutional neural network (CNN) with gradient-weighted class activation mapping (Grad-CAM) for spectral data analysis and model interpretability.

Main Results:

• The Grad-CAM-CNN model achieved 99% accuracy in predicting withering degrees.
• Demonstrated superior performance compared to conventional models.
• Achieved high computational efficiency (0.0559 s/100 samples).

Conclusions:

• The developed intelligent platform accurately predicts black tea withering degrees, ensuring aroma quality.
• The approach offers an efficient, scalable, and cost-effective solution for real-time tea manufacturing control.
• Potential exists for adapting this technology to other food systems.