Aim: Green coffee processing, before the roasting phase, requires effective removal of foreign materials and defective kernels to ensure product quality, process safety, and compliance with industrial requirements. The aim of this research is to use conventional RGB-based optical sorters for product sorting. These rely primarily on surface colour characteristics and can be limited when contaminants display visual similarities to healthy beans. Methods: Hyperspectral imaging (HSI) provides a non-destructive alternative by integrating spatial and spectral information in the visible and near-infrared (VIS/NIR) range. In this study, a VIS/NIR HSI system was integrated into a commercial industrial optical sorter and validated under real operating conditions. Contaminated green coffee batches (10 kg) containing known amounts of organic and inorganic contaminants were processed through multiple sorting passes using a statistical classification logic embedded into the sorter programmable logic controller (PLC) for real-time decision making. Results: The system achieved complete removal of stone contaminants after a single pass, while organic contaminants (peel and defective beans) were substantially reduced across successive cycles. After two sorting passes, the cumulative yield of compliant coffee beans was approximately 84%, representing an acceptable trade-off between contaminant removal efficiency and product loss in an industrial context. Conclusions: Overall, the results support the feasibility of deploying VIS/NIR hyperspectral sensing for high-throughput industrial coffee sorting, with potential advantages in discrimination capability compared with conventional colour-based systems.
Berardi et al. (Mon,) studied this question.