Hyperspectral imaging combined with data fusion for non-destructive prediction of soluble solids content and water-soluble sugars in strawberries during storage after static-pressure bruising.

Soluble solid content (SSC) and water-soluble sugar contents are key indicators for evaluating strawberry quality. Static-Pressure Bruising can alter these components, thereby affecting fruit texture and overall eating quality. This study aimed to investigate the feasibility of predicting SSC as well as fructose and glucose contents in strawberries subjected to Static-Pressure Bruising followed by different storage durations using two hyperspectral imaging techniques (Vis-NIR and NIR) combined with low-level and mid-level data fusion strategies (LLF and MLF). After applying five spectral preprocessing methods and three characteristic wavelength selection approaches, two conventional machine-learning models and four deep-learning models were developed. The results showed that the data fusion strategies outperformed single-spectrum models for all prediction targets. The optimal SSC prediction model was LLF combined with IVSO-LSSVM, achieving an Rp of 0.8930, an RMSEP of 0.4982°Brix, and an RPD of 2.2312. For fructose prediction, the best model was MLF combined with IVSO-TCN-BiGRU-Self-Attention, with an Rp of 0.7691, an RMSEP of 3.8737 mg/g, and an RPD of 1.5707. For glucose prediction, the optimal model was LLF combined with IVSO-LSSVM, yielding an Rp of 0.7435, an RMSEP of 3.0028 mg/g, and an RPD of 1.5014. Based on the optimal models, visualization maps of SSC and water-soluble sugar distributions in strawberries were generated. These results demonstrate the feasibility of hyperspectral imaging combined with data fusion strategies for monitoring quality variations in strawberries subjected to static pressure bruising, although further optimization is needed to achieve accurate quantitative performance for individual sugars.