Improving classification accuracy of spectrally similar tree species: a complex case study in the Kruger National Park

Abstract

This presentation to the SASA Eastern Cape Chapter at Rhodes University studies the variability within a species class and the variability between the species classes of seven spectrally similar tree species and presents ways in which the within-species class variability can be reduced compared to the between-species class variability. Furthermore, two classification approaches with spectral angle mapper: (i) using a spectral library composed of one spectrum (endmember) per species and (ii) a multiple endmember approach is presented, conventionally called K-nearest neighbour classifier.