ORBMO-RF: a non-destructive classification method for ginseng seeds based on multimodal fusion and improved red-billed blue magpie optimization algorithm.

Introduction: Ginseng, as a precious medicinal plant, requires precise classification of its seeds, which directly impacts production processes and the stability of herbal quality. Furthermore, this classification plays a critical role in advancing ginseng breeding and the modernization of the industry. Current research indicates that systematic automated precision classification technologies for ginseng seeds remain underdeveloped, necessitating breakthroughs in technical bottlenecks. Methods: This study innovatively proposes a smart classification method based on multimodal data fusion. It employs recursive feature elimination (RFE) to select morphological features from images, followed by competitive adaptive reweighted sampling (CARS) to extract spectral bands from hyperspectral data within the 350~2500 nm range. Morphological and spectral features are then integrated to construct a random forest (RF) classification model optimized using an enhanced, red-billed blue magpie optimization (RBMO) algorithm. To address the RBMO algorithm's tendency to converge to local optima, the hybrid optimization framework is constructed by integrating three mechanisms: the improved Circle chaotic map, the golden sine search strategy, and the adaptive simulated annealing perturbation mechanism. Results: Experimental results demonstrate that the proposed model outperforms the baseline model RF, achieving 4.69%、4.79%、4.69 and 4.74% improvements in classification accuracy, precision, recall, and F1-score on test datasets, respectively. Discussion: The established multimodal data fusion classification system not only provides theoretical and technical foundations for industrial-scale ginseng seed classification but also offers a transferable intelligent decision-making paradigm for non-destructive testing in traditional Chinese medicine.