Classification of ginseng year based on visible-near infrared hyperspectral images combined with image segmentation

Ginseng can regulate the immune system and enhance the body's resistance. Conventional identification methods relying on morphological features or chemical assays are often destructive, time-consuming, and subjective. This paper focuses on the non- destructive and rapid detection of ginseng's years. Compared with traditional methods, non- destructive detection method can maintain the integrity of the samples, which is conducive to subsequent research. The method can reduce the detection cost and enhance the detection efficiency. Visible/infrared hyperspectral imaging data (wavelength range: 400–1700 nm) were collected from 25 ginseng samples across five different aging years (8, 10, 15, 20, and 24 years). A hyperspectral imaging analysis combined with an image segmentation was developed for the rapid and non-destructive classification of ginseng samples by five different years. After preprocessing the spectral data, classification models were established using partial least squares discriminant analysis (PLS-DA), support vector machines (SVM) and random forest( RF). The accuracy of prediction results was used as an evaluation standard to choose the optimal model, while model performance was further assessed using root mean square error(RMSE) and confusion matrix analysis. The PLSR regression model was excluded due to poor discriminative ability (RMSE = 0.89034). SVM achieved a prediction accuracy of 68.4%, PLS-DA achieved a prediction accuracy of 94.4%, and RF achieved a prediction accuracy of 87.6%. The prediction accuracy of PLS-DA consistently exceeded 92%, and that of RF exceeded 84%. The graph segmentation algorithm combined with hyperspectral images provides an accurate, and non-destructive method for precise identification of ginseng year.