Combining VNIR and NIR hyperspectral imaging techniques with a data fusion strategy for the determination of fat...

Fat content and acid value are critical indicators for evaluating walnut quality. This study employs two hyperspectral imaging techniques (visible near-infrared (VNIR) and NIR) in combination with low-level and mid-level fusion strategies (LLF and MLF) for the prediction of these indicators across different storage periods and classify walnuts accordingly. Prediction models, including partial least squares regression (PLSR), particle swarm optimization-support vector regression (PSO-SVR), and random forest (RF), were developed after preprocessing and feature wavelength selection. The results showed that the data fusion strategy exhibited better performance in both indicators compared to individual data. The MLF strategy was best in predicting fat content based on RF combined with uninformative variable elimination (UVE), achieving an R_p² of 0.8706, an RMSEP of 0.0083, and an RPD of 2.7797. The LLF strategy showed optimal performance in predicting acid value based on PSO-SVR combined with uninformative variable elimination-competitive adaptive reweighted sampling (UVE-CARS), achieving an R_p² of 0.9694, an RMSEP of 0.0369, and an RPD of 5.7202. Storage period classification achieved 100 % accuracy for walnuts stored for 6 and 18 months using both VNIR and NIR spectral data. These findings highlight the great potential of hyperspectral imaging in combination with data fusion for rapid, nondestructive walnut quality assessment and storage period identification.