Abstract: Representative tobacco leaf samples (orange-yellow, lemon-yellow, and red-brown) were collected and preprocessed. Factor analysis was employed to extract features related to tobacco leaf color, reducing data dimensionality and capturing key information. The AdaBoost algorithm was used to construct a classification model for the extracted features, and its predictive performance was compared with other algorithms. The performance of the FA-AdaBoost model was evaluated, and its classification effectiveness was validated. The results showed that three wavelengths (380, 460 nm, and 740 nm) were selected as key spectral features for tobacco leaf color classification using factor analysis. Compared with gradient boosting, Bagging, and random forest algorithms, the AdaBoost algorithm achieved the lowest test error rate with fewer iterations. The FA-AdaBoost model demonstrated excellent performance in tobacco leaf color classification, with high precision, recall, and F1-score. The FA-AdaBoost model achieved remarkable recognition results for red-brown leaves, with all three metrics reaching 100%. In terms of support, significant differences in sample sizes across categories were observed, with red-brown leaves (3 samples) being substantially fewer than other categories, indicating evident class imbalance. Nevertheless, the FA-AdaBoost model achieved an overall accuracy of 86%, demonstrating its strong classification capability despite class imbalance challenges. The AdaBoost algorithm exhibited efficient and accurate recognition in tobacco leaf color classification tasks, with balanced performance across different categories and robust generalization ability.

Key words: tobacco grading, factor analysis, AdaBoost algorithm, tobacco leaf color, classification