Semi-automated procedures for tree species identification in high spatial resolution data from digitized colour infrared-aerial photography

Abstract

A semi-automated classification procedure is presented in this paper for identification of forest species from digitized large-scale, colour-infrared (CIR) aerial photographs to simulate imagery from future sensors with high spatial resolution capability (below 0.5 m). The applied computer-assisted classification approaches involving a tree by tree approach consist of basically four steps: the digitization of crown shapes (polygons); the actual classification within these polygons; the determination of the most frequent class within the polygons; and the filling of the polygons with this particular class. The best result was achieved with a parallelepiped classifier involving the original near-infrared channel, a texture feature (standard deviation), and four features produced with a principal component/colour enhancement. An average of about 80% of the trees could be correctly classified.