Food grading has always been a research topic because of large variations among food products.
Many subjective assessment methods with poor repeatability and tedious procedures are still widely used. In
this study, a hyperspectral-imaging-based technique was developed to achieve fast, accurate, and objective
potato grading. The system was able to extract the morphological features and spectral responses on water
content in potatoes simultaneously. A significant feature wavelength range (934-997 nm) was found to be a
sensitive water absorption band for predicting the water content in potato samples. Artificial Neural Network
was engaged to establish the water content prediction model. The results showed that the R between the
predicted and actual water content was 0.932 and 0.769 for training and validation, respectively. The rootmean-
squared-error was found to be less than 0.014 for both training and validation. The weights of the
potatoes were predicted based on two indices, one image-based index (1) and another index (2) including
water content information. The prediction errors with index 2 was much less than that with index 1. Hence,
combining morphological features and spectral responses, 2 the weight measurement for potatoes could be
improved.