Prediction of apple internal qualities using hyperspectral imaging techniques

摘要

Sugar content and firmness are significant internal properties and have important significance for quality grade and industry production. The hyperspectral imaging techniques was investigated for prediction of sugar content and firmness in apples. Mean spectra were extracted from the regions of interest of the image between 400nm and 1000nm. After multiplicative scatter correction (MSC) preprocessing, the regression coefficient of partial least squares (PLS) was used to extract feature wavelengths. 6 feature wavelengths of sugar content (384.2, 418.9, 502.5, 675.2, 720.9, 968.7nm) and 6 feature wavelengths of firmness (418.9, 500.0, 513.5, 599.9, 704.4, 964.7nm) were extracted. The performance of prediction model of sugar content and firmness in full wavelength and feature wavelength was compared, 3 prediction models like BP artificial neural network (BP), partial least squares regressions (PLSR) and principal component regression (PCR) model of sugar content and firmness were built, respectively. The results showed that the sugar content and firmness can be predicted by 3 types of models. The BP model based on feature wavelength achieved the optimal performance for sugar content and firmness. The correlation coefficient and root mean square errors of prediction by BP were 0.9225, 0.137 for SSC and 0.9126, 1.099 for firmness, respectively. The study demonstrated that hyperspectral imaging technique can be a reliable tool to detection of apple internal qualities, which provided a theoretical reference for nondestructive detection of apple.