An innovative multivariate strategy for HSI-NIR images to automatically detect defects in green coffee.

In the present study, an advanced and original multivariate strategy for the processing of hyperspectral images in the near-infrared region is proposed to automatically detect physico-chemical defects in green coffee, which are similar one to each other by naked eye. An object-based approach for the characterization of individual beans, rather than single pixels, was adopted, calculating a series of descriptive parameters characterizing the distribution of scores on the lowest-order principal components. On such parameters, the k-nearest neighbors (k-NN) classification algorithm was applied and the predictive results on the test samples indicate that this approach is able not only to distinguish defective beans from non-defective ones, but also to differentiate the various types of defects. Hyperspectral imaging is demonstrated to be a valid alternative for the sorting of green beans - a crucial phase for coffee import/export.