Automatic Segment-Level Tree Species Recognition Using High Resolution Aerial Winter Imagery

Our objective was to automatically recognize the species composition of a boreal forest from high-resolution airborne winter imagery. The forest floor was covered by snow so that the contrast between the crowns and the background was maximized. The images were taken from a helicopter flying at low altitude so that fine details of the canopy structure could be distinguished. Segments created by an object-oriented image processing were used as a basis for a linear discriminant analysis, which aimed at separating the three dominant tree species occurring in the area: Scots pine, Norway spruce, and downy birch. In a cross validation, the classification showed an overall accuracy of 81.9%, and a kappa coefficient of 0.73.