FEATURESELECTIONFOR MORPHOLOGICAL FEATURE EXTRACTIONUSINGRANDOM FORESTS
Sveinn R. Joelsson, Jón Atli Benediktsson · 2006
Morphologicalfeature extraction (MFE) hasbeensuccessfully usedtoincrease classification accuracy andreduce thenoise level forclassification ofaerial images. Inthis paperweexplore feature selection andextraction forMFE (a) using random forests (RFs) forclassification andfeature selection. Theapproach iscompared toMFEfrom principalcomponents extracted from thedata, byprincipal component analysis (PCA), whichhasbeensuccessful inthe past. Theexperimental results presented inthis papershow thatbyestimating themostimportant features ofourdata setusing RFs,andselecting afewofthefeatures forMFE yields equal orbetter accuracies thanbyusing PCs. (b)