Determination of impurities in grinded coffee using near infrared hyperspectral imaging

Coffee is one of the most consumed beverages worldwide, ranked only after water. Especially in Brazil, coffee is part of the country history and economic development. Indeed, the Brazilian Coffee Industries Association (ABIC) has a strong quality program to maintain a high level quality of the product sold. The brands can be granted with a “Quality Symbol”, which describes the overall quality of the product, and with a “Purity Stamp”, which ensures the coffee is free of impurities. The only impurities that are allowed are husks and branches that are residual from the harvest and unpeeling process, but their concentration is limited to 1%. Any other impurity, such as sugar, corn, starch or barley, is unacceptable and it is considered as adulteration. The method used for roasted and ground coffee inspection is based on visual investigation using optical microscopy, which is an empiric method that deeply depends on the analyst experience. Moreover, the sample preparation requires the use of a high volume (60 mL for each sample) of chlorinated solvent, filtering, drying and the identification and separation of husks and branches is manually made with the help of a stereomicroscope. Despite the long time needed, many errors can be introduced during this process, mainly personal (subjective) errors.
The aim of this work was to develop an objective procedure using near infrared hyperspectral imaging spectroscopy (HI-NIR) and chemometrics to detect impurities in coffee samples without sample preparation.
Five different percentages ranging from 0.5 % to 20 % of husks and branches were added to pure coffee and the samples were submitted to optical microscopy and to HI-NIR analysis. In total 22 samples have been measured including samples of pure coffee and pure husks. Hyperspectral images were collected using an NIR hyperspectral line scan imaging system (Specim, Finland). Each image consisted of 256-pixel lines acquired at the 1,100-2,400 nm wavelength range. Spatial resolution was set to 150 µm.
A simple unsupervised PCA model has been developed using HI-NIR spectra of pure products, i.e. coffee and husks. The idea was to characterize coffee based on the two first principal components, which showed a clear discrimination between pure coffee and husks. Scores of the images on the mixtures to be predicted are projected into the space defined by those PCs and the number of pixels detected as husks are quantified and used to detect the level of contamination. Compared to the reference method, where the recovery of husks was below 20%, the preliminary results obtained with Hyperspectral Imaging combined with PCA indicated the utility of the method to detect impurities even at values of 0.5 % of husks. However, quantification remains a challenge and a correlation between the number of pixels detected as husks and the real percentage of the impurity should be defined.