Rapid identification of Chrysanthemi Flos based on multi-source data fusion and machine learning algorithms.

Area of Science:

• Integrative Medicine
• Cheminformatics
• Computer Vision

Background:

• Chrysanthemi Flos (CF) has visually similar varieties, complicating identification.
• Current methods often rely on single traits, lacking a holistic approach.
• A systematic method linking morphology to chemical profiles in Traditional Chinese Medicine (TCM) is needed.

Purpose of the Study:

• To develop a data fusion strategy integrating computer vision and chemical analysis for CF cultivar identification.
• To investigate the correlation between macroscopic phenotypic features and bioactive constituents in CF.
• To establish precise identification models for CF cultivars.

Main Methods:

• Collected 101 CF samples across seven cultivars.
• Extracted Red-Green-Blue (RGB) and Gray-Level Co-occurrence Matrix (GLCM) texture features using Python.
• Quantified seven flavonoid and phenylpropanoid components via ultra-high performance liquid chromatography (UPLC).
• Applied Random Subspace Method (RSM), Wide Neural Network (WNN), Support Vector Machine (SVM), and K-Nearest Neighbor (KNN) for cultivar identification.

Main Results:

• RGB features correlated positively with chlorogenic acid and apigenin-7-O-glucoside.
• GLCM texture features correlated positively with 3,5-O-dicaffeoylquinic acid.
• All four identification models showed high accuracy; RSM achieved 94.4% (training) and 96.7% (test).

Conclusions:

• Macroscopic visual phenotypes of CF are inherently linked to their phytochemical profiles.
• The developed multi-source data fusion framework enables fast and precise identification of CF cultivars.
• This study provides a data basis and technical scheme for TCM quality appraisal.