Using a simple digital camera and SPA-LDA modeling to screen teas

Abstract

Classification or screening analysis of natural unprocessed teas using simple digital images and a variable selection algorithm is described. The proposed methodology uses color histograms generated on free downloadable software ImageJ 1.44p as a source of analytical information. Two chemometric methods were compared for classification of the resulting images, namely Soft Independent Modeling of Class Analogy (SIMCA), and Linear Discriminant Analysis (LDA) with variable selection by the Successive Projections Algorithm (SPA). The results were evaluated in terms of errors found in a sample set separate from the modeling process. The choice of more informative photometric color attributes (red-green-blue (RGB), hue (H), saturation (S), brightness (B), and grayscale) for screening the tea samples was made during the color modeling because SIMCA failed to give good results. Therefore the data treatment used SPA-LDA, which correctly classified all samples according to their geographical regions, whether from Brazilian, Argentinian or foreign soils.