Rapid identification of Fritillaria spp. using multi-wavebands spectroscopy and multi-source data fusion strategies

Abstract

Fritillaria spp., a kind of typical medicine food homology plant, is in growing demand for rapid identification due to its diverse types. This study aims to utilize the mid-infrared (MIR) and ultraviolet-visible-near infrared (UV–VIS–NIR) spectroscopy for the first time to identify Fritillaria spp. Two machine learning techniques of random forest (RF) and support vector machine (SVM) are employed to establish classification models based on spectral data from various Fritillaria spp. Data fusion strategies including low-level, mid-level and high-level fusions are applied to combinate MIR and UV–VIS–NIR data. The result indicates that data fusion improved the classification of Fritillaria spp. compared to using individual spectral data. The RF classifier showed superior performance over the SVM model. The high-level data fusion model achieved the highest prediction accuracy of 93.31 %. This study demonstrates that integrating MIR and UV–VIS–NIR spectroscopy with data fusion techniques offers a feasible, non-destructive, and rapid approach for classifying Fritillaria spp., highlighting the potential for the identifying related medicinal food homology products.