Intelligent species identification of Fritillaria spp. based on multispectral feature fusion

Fritillaria spp. refers to the dried bulbs of multiple Fritillaria species (Liliaceae family), a well-known medicinal and edible homologous plant widely used in traditional herbal medicine and dietary supplements. Different Fritillaria spp. exhibit differences in chemical compositions, flavors, properties, and medicinal values, leading to a growing demand for quick authentication. This study was to explore the feasibility of rapid identification Fritillaria spp. using portable near-infrared (NIR), mid-infrared (MIR) and Raman spectroscopy coupled with machine learning methods. Classification models using K-nearest neighbors (KNN), random forest (RF), and support vector machine (SVM) were developed based on spectral data collected from four series of Fritillaria spp. The data was preprocessed by multiple data pretreatment and feature variable selection algorithms to improve its quality. Low-level, mid-level, and high-level data fusion strategies were applied to combine the spectral data from portable NIR, MIR, and Raman techniques. The results showed that data fusion significantly improved the classification of Fritillaria spp. Notably, high-level fusion based on the stacking strategy yielded the highest prediction accuracy at 90.41 %, with an F1 score of 0.906, a Kappa value of 0.861, and an AUC of 0.989. This study provides a rapid, accurate, and non-destructive method for identifying Fritillaria spp. samples, thereby enhancing analytical capabilities for future research related to this species.