Application of Hyperspectral Imaging for Prediction of Textural Properties of Maize Seeds with Different Storage Periods

In this study, hyperspectral imaging in the spectral range of 400–1,000 nm was applied for predicting texture changes of maize seeds by different storage times, mainly including hardness, springiness, and resilience. Before modeling, a novel denoising algorithm, namely orthogonal signal correction (OSC), was developed for spectral pre-treatment. Then, partial least squares regression (PLSR) algorithm was used to integrate the extracted spectral data with the reference values measured by a traditional method. The established PLSR models yielded good results for predicting hardness, springiness, and resilience (R 2 P = 0.9012, RMSEP = 0.0502; R 2 P = 0.8744, RMSEP = 0.0038; R 2 P = 0.8477, RMSEP = 0.0057, respectively). In order to simplify the prediction models, successive projection algorithm (SPA) was employed to choose the most important wavelengths that had the greatest influence in predicting the textural properties. As a result, seven (400, 416, 450, 501, 517, 530, 980 nm), seven (400, 403, 420, 525, 552, 749, 835 nm), and six (436, 482, 532, 700, 735, 939 nm) optimal wavelengths were selected by SPA for hardness, springiness, and resilience prediction, respectively. Based on the optimal wavelengths, new OSC–SPA–PLSR models were built and showed good results for predicting hardness, springiness, and resilience of maize samples with high R 2 P of 0.8365, 0.8217, and 0.7930, respectively, and relatively low RMSEP of 0.2085, 0.0530, and 0.0595, respectively. Due to that no more than seven wavelengths were needed, the OSC–SPA–PLSR models were used for visualization of the values of textural properties in maize seeds as affected by different storage times. The results showed that hyperspectral imaging technique could be used as a fast, accurate, and non-destructive tool for predicting textural properties of maize as affected by different storage times.