Integration of algorithm optimization and interpretability: Non-destructive age identification of forest-cultivated dried ginseng using hyperspectral imaging

This is the authors' abstract. We don't add key points for this paper.

Forest cultivated dried ginseng is a valuable medicinal material whose cultivation age influences its quality and market value, yet conventional identification methods are often destructive or inefficient. In this study, hyperspectral imaging (HSI) combined with a Random Forest model optimized by Improved Holistic Swarm Optimization (IHSO-RF) was employed for the nondestructive discrimination of samples aged 5, 6, and 7 years. Among nine preprocessing methods, Savitzky-Golay smoothing combined with multiplicative scatter correction (SG+MSC) provided the best performance. CARS and RFE were applied to extract informative spectral wavelengths and textural features, respectively, for feature fusion. The proposed IHSO framework integrated cascaded Logistic Tent chaotic initialization, an adaptive polynomial exponential step size strategy, and adaptive Lévy flight with sinusoidal mutation to enhance population diversity, balance global exploration and local exploitation, and improve convergence performance. The Fusion-IHSO-RF model achieved the best classification performance, with accuracy, precision, recall, and F1-score values of 0.9489, 0.9610, 0.9622, and 0.9616, respectively. SHAP analysis indicated that spectral features contributed strongly to age discrimination, while texture features provided complementary information. This method enables accurate non-destructive classification of adjacent cultivation ages of forest-cultivated dried ginseng, providing an effective strategy for quality grading and market standardization of traditional Chinese medicinal materials.