NIR hyperspectral imaging for predicting the composition of granular food commodities

Hyperspectral imaging (HSI) in the Near-Infrared (NIR) spectral range was applied for non- destructive characterisation of three staple food commodities: wheat, cocoa and coffee. Industrially-relevant properties such as moisture, fat and proteins were explored on a single seed basis. Prediction models were built for whole wheat kernels, cocoa seeds (de-shelled i.e., cotyledons or nibs) and green coffee beans. Major constituents were successfully predicted in the three commodities with performance allowing quantitative prediction for screening purposes and quality control. In addition, chemical compounds found at lower concentrations were analysed. This comprised indirect methods for enzymatic activity in wheat, polyphenols and antioxidant activity in cocoa, and sucrose, caffeine and trigonelline in green coffee beans. Calibration models built from HSI scanning of green and roasted coffee beans demonstrated the potential to predict generated volatile compounds upon roasting. This approach has been also performed to demonstrate the potential to understand variability at single kernel/seed basis, which can be used for quality improvement of food grains/seeds. HSI-based quantification for single seeds (as well as single pixel level) could be used as a selection tool to create different streams, e.g. for specific product characteristics or to obtain a more consistent composition of the final food product or segregating materials into different process streams with different commercial values. This research work is of strong practical interest, due to the potential applications