Tree Species Classification and Input Data Evaluation

Abstract

Semantic world models require detailed classifications that allow the generation of entities that represent real world objects. This study combines advanced binary support vector machine classifiers in the hierarchical structure of a binary decision tree. This approach fulfills the requirements of a subsequent semantic model generation approach as it is an object based approach that works on the single tree level. It is fast and can be applied easily without expert knowledge. The algorithm can be trained on existing tree sample inventory data and additional samples from other data sources can be imported or placed. Classification results were calculated on a variety of input data sources and spatial resolutions. The achieved accuracies were analyzed and provide information for the decision support regarding input data choice for future applications.