Application of a pattern recognition algorithm for single tree detection from LiDAR data

Abstract

In the present study, we applied the Particle Swarming Optimization (PSO) procedure to parametrize two Local Maxima (LM) algorithms and a pattern recognition model based on raster and point-cloud datasets in order to extract treetops of coniferous forests from high resolution LiDAR-data of different forest structures (monoplane, biplane and multi-layer) in the Alps region. The approach based on the pattern recognition model uses the geomorphon algorithm applied to the DSM to detect the treetops. The geomorphon model gave good results in terms of matching rates (Rmat: 0.8) with intermediate values of commission and omission rates (Rcom: 0.22, Rom: 0.2). Therefore, it could be a valid alternative to the LM-algorithms when only raster products (DSM - CHM) are available. The geomorphon pattern recognition model has been proved to be a powerful method in order to properly detect treetops of coniferous stands with complex forest structures. This model allows to obtain high detection rates and estimation accuracy of forest volume, also in comparison to the most recent available literature data. The models are developed in Java under Free and Open Source license and are integrated in the JGrassTools library, which is now available as SpatialToolbox of the GIS gvSIG.