Comparative study of portable Vis-NIR spectrometers for corn moisture content prediction using machine learning

The non-destructive estimation of the corn kernel moisture content is essential for determining the optimal harvest period. Although various spectrometer sensors are currently available, their predictive performance differs due to variations in the spectral resolution and wavelength coverage. This study compared the performance of several portable spectrometer sensors with different wavelength ranges for predicting the corn moisture content. Spectral data and reference moisture content were used to develop the prediction models using partial least squares regression (PLSR) and an artificial neural network (ANN). Based on the PLSR modelling, the AS7265X and C12880MA sensors produced the best performance, with coefficients of determination (R²) for training and testing reaching up to 0.90. Furthermore, the ANN modelling yielded improved predictive accuracy, with the highest R² value of 0.95 obtained using the same sensor combination. These results demonstrate that portable spectrometers show strong potential for the non-destructive field-based prediction of the corn moisture content and can serve as a reliable indicator for determining the optimal harvest timing.