Abstract
The lily, valued for its edibility and medicinal properties, is rich in essential nutrients. However, storage conditions and sulfur fumigation during processing can degrade key nutrients like polysaccharides, phenols, and sulfur dioxide. To address this, we applied a deep learning model combined with hyperspectral imaging for the rapid prediction of nutrient quality. The CLSTM (convolutional neural network-long short-term memory) model, utilizing variable combination population analysis (VCPA) for wavelength selection, effectively differentiated sulfur fumigation patterns in lilies. In terms of nutrient content prediction, the CLSTM model combined with full-wavelength data demonstrated superior performance, achieving an R(2) value of 0.769 for polysaccharides and 0.699 for total phenols. Additionally, the CLSTM model combined with IRF-selected characteristic wavelengths exhibited remarkable performance in predicting sulfur dioxide content, with an R(2) value of 0.755. These findings highlight the potential of hyperspectral imaging and the CLSTM model in enhancing the quality assessment and ensuring the nutritional integrity of lily products.