Tea grading, blending, and matching based on computer vision and deep learning.

Area of Science:

• Agricultural Science
• Computer Science
• Food Science

Background:

• Traditional tea assessment methods lack efficiency and accuracy.
• Computer vision and deep learning offer solutions for tea quality control.
• Developing non-destructive methods is crucial for modern tea production.

Purpose of the Study:

• To develop an efficient, non-destructive method for tea grading, blending ratio evaluation, and sample matching.
• To enhance tea quality assessment using deep learning and attention mechanisms.
• To improve the accuracy and speed of tea analysis in production.

Main Methods:

• Trained a Residual Network (ResNet) model on an enhanced tea image dataset.
• Integrated Convolutional Block Attention Module (CBAM) to boost feature extraction.
• Utilized deep learning for image analysis in tea quality evaluation.

Main Results:

• Achieved 95.05% accuracy for oolong tea grade classification and 99.13% for black tea.
• Outperformed other deep learning models like EfficientNet, MobileNet, and VGG16.
• Demonstrated high efficiency in oolong tea blend evaluation (2.26% error) and black tea sample matching (3.34% error).

Conclusions:

• Attention mechanisms are vital for analyzing intricate image textures in tea.
• Deep learning with attention modules significantly enhances tea quality evaluation accuracy and efficiency.
• Intelligent classification methods can modernize tea production, ensuring consistent quality.