Ensemble of predictors and laser induced breakdown spectroscopy for certifying coffee

• Published in: Electronics letters Vol. 47; no. 17; p. 1
• Main Authors: Ferreira, E J, Ferreira, E C, Delbem, A C B, Milori, D M B P
• Format: Journal Article
• Language: English
• Published: Stevenage John Wiley & Sons, Inc 18.08.2011
• Subjects: Brazil; Classification; Coffee; Decision trees; Laser induced breakdown; Pattern recognition; Spectra; Spectroscopy
• ISSN: 0013-5194

The commercial demand and the economic value of roasted and ground coffee (RGC) are intrinsically related to its quality. Brazilian RGC has been certified for quality based on a score assigned by expert tasters. However, intrinsic costs and the lack of fast techniques make large-scale analysis prohibitive. Laser induced breakdown spectroscopy (LIBS) is a promising technique for RGCs, however, it demands an accurate coupled-model for pattern recognition. In this reported work, two ensemble methods in ranges around 680 nm of LIBS spectra are employed to construct ensembles of decision trees for RGC classification. Results show accuracies above 84% for ensemble methods in some spectral ranges. The best performances were in the range 670 nm-690 nm, making the design of dedicated equipment with low-cost devices viable.