Selecting a representative training set for the classification of demolition waste using remote NIR sensing

In the AUTOSORT project, the goal is the separation of demolition waste in three fractions: wood, plastics and stone. A remote near-infrared sensor me asures reduced reflectance spectra (mini-spectra) of objects. Linear discri minant analysis (LDA) is used for the classification of these spectra. To o btain the LDA model, a representative training set is needed. New LDA-model s will be regularly needed for recalibrations. Small training sets will sav e a lot of labour and additional costs. Two object selection methods are investigated: the Kennard-Stone algorithm and a statistical rest procedure. Training sets are acquired from which the mini-spectra are used to obtain LDA models. In the training sets, the obje ct amounts and their ratios are varied. Two object ratios are applied: the ratios as they occur in the complete data set and the equalised ratios. The Kennard-Stone selection algorithm is the preferred method. It gives a u nique list of objects, mainly sampled at the cluster borders: partial clust er overlap is better defined. This is in contradiction with the sets of obj ects, accepted by the statistical test procedure: those objects tend to occ ur around the fraction means. This is a drawback for the classification per formance: some accepted training sets are unacceptable. The ratios between the fraction amounts are not important, but equal fraction amounts are pref erred. Selecting 25 objects for each fraction should be suitable. (C) 1999 Elsevier Science B.V. All rights reserved.