Moisture content prediction of cigar leaves air-curing process based on stacking ensemble learning model

Area of Science:

• Agricultural Engineering
• Image Processing
• Machine Learning

Background:

• Accurate moisture content determination is vital for preserving cigar leaf quality during air-curing.
• Traditional methods are often subjective, destructive, and impractical for real-time monitoring.

Purpose of the Study:

• To develop a non-destructive, image-based stacking ensemble learning model for predicting cigar leaf moisture content.
• To evaluate the model's performance and identify key predictive features.

Main Methods:

• Collected front and rear surface images of cigar leaves throughout the air-curing process.
• Extracted color and texture features, applied filtering, and used the entropy weight method for model selection.
• Constructed a stacking ensemble model (MLP, RF, GBDT base learners; LR meta-learner) and applied SHAP for feature contribution analysis.

Main Results:

• The stacking ensemble model achieved a high prediction accuracy (R ² test =0.989), outperforming traditional models.
• SHAP analysis indicated front surface features (45.5%) and leaf features (38.5%) were most influential.
• Key predictors included airing period (AP), a*_f, G_f, and ASM_f.

Conclusions:

• The developed model offers a feasible and scalable solution for real-time, non-destructive monitoring of cigar leaf moisture content.
• Provides effective technical support for similar agricultural and food drying applications.