Shortwave infrared hyperspectral imaging can predict moisture content in grafted cucumber seedlings

Abstract

This study aimed to develop statistical model for predicting the moisture content in grafted cucumber (Cucumis sativus L.) by analyzing shortwave infrared hyperspectral images of scions of cucumber seedlings in the range of 1,000-2,400 nm. These images were processed using brightness correction and two sets of data were created by normalizing the spectral reflectance and conducting the first derivative. Statistical analysis of the spectral reflectance and moisture content of the grafted cucumber scions was used to interpret the correlation coefficient spectrum, multiple linear regression, and partial least square (PLS) regression to select the significant wavelengths and develop a model that could predict moisture content. The normalized PLS regression model showed better prediction performance than the other analysis methods tested. The determination coefficient and the root mean square difference for the validation dataset were 0.66 and 1.25%, respectively. These results suggest that the moisture content of grafted cucumber seedlings can be predicted in a nondestructive manner using hyperspectral images with a proper model.