Development of a Generalizable IRIV‐ANN Model for Rapid, Non‐Destructive Polysaccharides Detection in Polygonatum Sibiricum

ABSTRACT Polysaccharides, as important bioactive components in medicinal and functional food materials, are key indicators for quality evaluation. Conventional analytical methods for polysaccharides quantification often require laborious sample preparation and may cause irreversible damage to samples, which limits their suitability for routine quality control. To overcome these limitations, we developed a rapid and non‐destructive detection approach by integrating hyperspectral imaging with an iteratively retained information variables‐artificial neural network (IRIV‐ANN) model. Spectral data (400–1000 nm) were processed using the IRIV algorithm, which classified variables into four categories‐strong information, weak information, no information, and interference. The optimal subset was then used to construct an ANN model, and its performance was compared with models based on shuffled frog leaping algorithm (SFLA), successive projections algorithm (SPA), one‐dimension convolutional neural network (1D‐CNN), full‐spectrum ANN, and multiple linear regression (MLR). The IRIV‐ANN model successfully eliminated redundant variables, reduced model complexity, and improved predictive performance, achieving a coefficient of determination (R 2 ) of 0.9301 and a root mean square error (RMSE) of 0.7348 for the prediction set. The characteristic wavelengths selected were closely associated with vibrational changes in O‐H and C‐H functional groups, providing chemical interpretability of the spectral features. This method provides an efficient and reliable analytical tool for rapid quantification of polysaccharides and has considerable potential for broader application in the quality control of functional foods, traditional medicines, and other plant‐derived materials.