An Intelligent Identification Method for the Processing Degree of Stir-Fried Cannabis fructus Based on Multi-Source Information Fusion and Machine Learning

Cannabis fructus (CF) has been used as a functional food raw material. The rapid identification of its processing degree is vital for ensuring consistency of product quality, achieving standardized processing, and maintaining market authenticity. This study established an intelligent identification method for the processing degree of stir-fried CF based on multi-source information fusion and machine learning. Color (L, a, b) and odor signals of the samples were collected through machine vision and electronic nose systems. The total contents of fatty oil and trigonelline were determined. Among the evaluated models, XGBoost demonstrated superior performance, achieving near-perfect training accuracy (93.62%) and test accuracy (91.67%). Shapley Additive Explanations analysis revealed that color and signals from specific odor sensors were key distinguishing features. The color exhibited a gradual transition to light yellow, and total fat and oil content showed a likely positive correlation with a and b values. The content of senna alkaloids tended to reduce as frying proceeded to greater intensity. This work preliminarily uncovered the correlations among color, gas characteristics, and quality, and could provide new insights for intelligent monitoring of the preparation procedure.