Tree isolation and species classification

To achieve a fully automated forest inventory based on high resolution colo r infrared aerial images, two major digital image analysis tasks have to be solved: tree isolation, and tree species classification. A prototype softw are system capable of the following functionality has been developed and im plemented in 1985-1988. 1) Isolation of individual trees in dense forest stands by a Vision Expert System, and 2) Recognition of the tree species by a neural network. This work is reviewed pointing out advantages and disadvantages of the appr oach. An important issue is how to cope with the question of different spat ial extensions of the objects (i.e., spatial scale). New experiments show t hat scale space theory is well applicable. However, due to the very complex nature of the problem, any successful tree recognition system will require to model and to use additional expert knowledge.