Vis-NIR hyperspectral imaging along with Gaussian process regression to monitor quality attributes of apple slices during drying

Abstract

Hyperspectral images (400–1700 nm) of apple slices during the hot-air drying were acquired. The fusion of spectral data and Gaussian Process Regression (GPR) successfully predicted vitamin C (R-squared = 0.93 and RMSE = 0.57 mg/100 g fresh-weigh), SSC (R-squared ≈ 0.99 and RMSE ≈ 2.47%), moisture content (R-squared ≈ 1 and RMSE = 0.89%), and shrinkage (R-squared ≈ 0.98 and RMSE = 3.66% for shrinkage). The chromaticity was likewise well predicted, however, GPR models failed to predict rehydration ratio and total phenolic content. As hyperspectral systems are expensive and computationally intensive, their possible substitution with multispectral systems was investigated by finding optimal wavelengths. In this context, 1450 nm and 980 nm were singled out by using a combination of filter-based, wrapper-based, and embedded wavelength selection algorithms. The corresponding prediction accuracies for vitamin C, SSC, moisture content, and shrinkage were almost as good as those of the full spectrum. In the case of chromaticity, it is suggested to use a color camera as most of the efficient wavelengths laid in the visible range. These results indicate potential replacement of hyperspectral imaging by much simpler and lower-cost imaging sensors by which the way will be paved towards the appearance of smart dryers.