Tree Species Classification using Semi-automatic Delineation of Trees on Aerial Images

Abstract

The purpose of this study was to develop a method for classifying tree species from remote sensing images by combining a semi-automatic pattern recognition technique and spectral properties of trees. Five stands in southern Finland were studied. Individual trees in the digital colour infrared (CIR) aerial photographs were segmented by a method based on the recognition of tree crown patterns at subpixel accuracy. The images were filtered with the Gaussian N-by-N smoothing operator and local maxima above a threshold level were segmented. The segments were classified into three tree species classes. The kappa coefficients for stands varied from 0.43 to 0.86 when the training data and test data were from the same aerial photograph. When training data from other photographs were used as reference data, the kappa coefficients ranged from 0.40 to 0.75. The method described provides an interesting approach for detecting tree species semi-automatically in digital aerial data. Keywords: BirchCir AerialImageForest InventoryNorway SprucePattern RecognitionScots Pine