Classification Techniques for Material Identification on Soil Science Images

Moacir P. Ponti Junior, DC, UFSCar
Nelson D. A. Mascarenhas, DC, UFSCar

Abstract. Three classifiers and multiple classifier combining rules are presented to identify materials in soil science images obtained from a CT scanner with different energies.

[ project home page | contact authors ]


Classification performance improvement with classifier combination.
Picture 1) Result of a Parzen classifier, the best individual classifier performance (Kappa = 0.850)
Picture 2) Result of a Mean Combiner of Parzen, K-Nearest Neigbours and Logistic Classifier outputs (Kappa = 0.899)