Learning-based Spectral Regression for Cocoa Bean Physicochemical Property Prediction

Abstract

Cocoa bean quality assessment is essential for ensuring compliance with commercial standards, protecting consumer health, and increasing the market value of the cocoa product. This work introduces a scalable methodology for evaluating the quality of cocoa beans by predicting key physicochemical properties from the spectral signatures of cocoa beans using learning-based regression models and a VIS-NIR spectrometer integrated on a conveyor belt. Ground-truth values were obtained through standardized laboratory analyses following commercial cocoa quality regulations. Models were evaluated on samples from Santander, Colombia and other regions, achieving high R2 scores across properties.