Abstract:

To address the issue of low efficiency in existing intelligent tobacco grading models, this study focused on digital images of tobacco plant populations. Taking 31 phenotypic parameters from four main categories, including RGB color model skewness parameters, Lab color model parameters, HSV color model parameters, and leaf texture parameters, as input variables, a tobacco plant population grade classification model F₁ based on the Bayesian classification algorithm was constructed. Furthermore, a core parameter-based classification model F₂, utilizing the comprehensive factor loading score method, was proposed and verified. The overall accuracy of model F₂ reached 82.24%, representing a 12.82% improvement compared to model F₁, and the accuracy of all five grade judgments exceeded 70%. This study provides an applied theoretical basis for the development of an efficient and practical intelligent tobacco purchasing system.

Key words: factor analysis, digital image composite phenotype, Bayesian classification, flue-cured tobacco grade, population grading, intelligent grading model