Tree species classification in mixed Baltic forest

The paper addresses solution of the specific application task, namely, classification of individual trees to 5 conifer and deciduous species in mixed Baltic forest, based on processing of airborne hyperspectral and LiDAR data. Description of instruments and software used for data acquisition and preprocessing, image processing approach, obtained classification results and developed software application is presented. The proposed approach includes initial determination of design (training) sets for each species of interest, creation of species' clusters by adding randomly selected trees from the whole analyzed forest area, and final classification of all trees using Bayes classifier designed on the basis of clusters' properties. Coordinates of individual trees were estimated by processing of LiDAR data not discussed here. It is shown that classification error rate down to 3% can be achieved in favorable conditions.