Predicting the Degree of Fresh Tea Leaves Withering Using Image Classification Confidence

Area of Science:

• Agricultural Engineering
• Computer Vision
• Food Science

Background:

• Ensuring fresh tea leaf quality requires rapid, non-destructive methods to assess wilting.
• Current methods may lack the precision needed for real-time processing adjustments.

Purpose of the Study:

• To develop an accurate and efficient model for detecting the withering degree of fresh tea leaves.
• To establish a reliable method for calculating moisture percentage and determining wilting levels.

Main Methods:

• A fresh tea withering degree detection model based on image classification confidence.
• Incorporation of Receptive-Field Attention Convolution (RFAConv) and Cross-Stage Feature Fusion Coordinate Attention (C2f_CA) modules.
• A weighted method combining confidence levels and moisture labels to calculate moisture percentage.

Main Results:

• The proposed model achieved a classification accuracy of 92.7%, improving detection accuracy by 0.156.
• Excellent predictive performance for moisture content with Rp=0.9983, RMSEP=0.006278, and RPD=39.2513.
• Outperformed traditional Partial Least Squares (PLS) and Convolutional Neural Network (CNN) methods.

Conclusions:

• The developed model provides accurate and rapid detection of tea leaf withering.
• Offers crucial technical support for online determination during tea processing.
• Enhances quality control in tea production through precise wilting assessment.