Hyperspectral reflectance imaging for water content and firmness prediction of potatoes by optimum wavelengths

This article presents a method for nondestructively determining water content and firmness of potatoes using hyperspectral imaging (HSI) in the visible near-infrared (VIS/NIR) and short-wave infrared (SWIR) bands. Potatoes were scanned to acquire their hyperspectral images. First derivatives (FD), Savitzky–Golay (SG) smoothing, standard normal variable (SNV) and multiplicative scatter correction (MSC) were used to process the spectral data. Competitive adaptive weighted sampling (CARS) was employed to extract the effective wavelengths. Prediction models were established using several algorithms. The SG-CARS-partial least-squares regression (PLSR) model presented the best performance in the VIS/NIR band; the corresponding R 2 P , root mean square error of the prediction set (RMSE P ), and residual predictive deviation (RPD) values for water content and firmness were 0.9219 and 0.9118, 0.0034 and 0.0640 Newton (N), and 2.5780 and 2.4353, respectively. In the SWIR band, the FD-CARS-PLSR prediction model performed best; the R 2 P values for water content and firmness were 0.9313 and 0.9317, respectively, the RMSE P values were 0.0025 and 0.0216 N, and RPD values were 2.7453 and 2.7531. This study confirmed the feasibility of the HSI technology for nondestructively determining water content and firmness of potatoes.