Green-Making Stage Recognition of Tieguanyin Tea Based on Improved MobileNet V3

Area of Science:

• Agricultural Science
• Computer Science
• Food Science

Background:

• The green-making stage significantly impacts Tieguanyin tea's unique aroma and flavor.
• Current methods for recognizing green-making stages rely on subjective sensory experience, which is inefficient and labor-intensive.

Purpose of the Study:

• To develop a lightweight, automatic recognition model (T-GSR) for accurate and objective identification of Tieguanyin tea green-making stages.
• To overcome the limitations of traditional sensory evaluation in tea processing.

Main Methods:

• Collected a comprehensive dataset of Tieguanyin tea images across various green-making stages.
• Applied image preprocessing techniques including multi-color-space fusion and morphological filtering.
• Enhanced the MobileNet V3 backbone with an adaptive residual branch, GELU activation, and an Improved Coordinate Attention (ICA) mechanism.

Main Results:

• The T-GSR model achieved a recognition accuracy of 93.38% and an F1-score of 93.33%.
• The model demonstrates superior performance and reduced complexity compared to the original MobileNet V3, with 3.025 M parameters and 0.242 G FLOPs.
• The proposed model effectively captures complex tea features for accurate stage identification.

Conclusions:

• The T-GSR model provides an effective solution for intelligent recognition of Tieguanyin tea green-making stages.
• This facilitates online monitoring and supports the automation of tea production processes.
• The study highlights the potential of AI in enhancing traditional agricultural practices.