Application of optimized near-infrared spectral analysis for predicting moisture content and determining the origin of robusta coffee beans

Abstract

Moisture content is a critical quality attribute of green coffee beans because it directly affects storage stability, microbial safety, and final beverage quality. This study evaluated the effects of spectral preprocessing and wavelength selection on moisture prediction in green Robusta coffee beans from different geographical origins. It was investigated whether moisture-related wavelengths could be used for classification of origin. The results showed that combined preprocessing techniques, particularly multiplicative scatter correction or standard normal variate followed by the second derivative, improved predictive performance. The full spectra models achieved an R²_P of 0.95 and an RPD of 4.5 in prediction. The performance of the optimized eight wavelengths was an R²_P of 0.95 and an RPD of 4.2, very close to that of the full spectra models. Furthermore, the selected wavelengths retained useful information related to geographical origin. Cluster analysis based on both the full spectrum and the optimized wavelengths found two distinct clusters of coffee-growing regions. The classification accuracy based on the optimized wavelengths was 83.5%, while the accuracy reached 92.9% without Gia Lai samples. The results suggest that the optimized wavelengths selected for moisture prediction can be used for preliminary screening of the geographical origin of green Robusta coffee beans.