Segmentação de copas individuais de árvores nativas da transição cerrado-amazônia em nuvem de pontos

Abstract

The segmentations of individual trees in a pointcloud is still an unknown process for surveys in natural forests from the intertropical zone, thus complicating the acquisition of metric informations from the vegetation and surveys using laser scanning. The objective here was to evaluate a methodology for segmentating individual tree crowns on a pointcloud from the Cerrado-Amazônia ecotone. To achieve that, LiDAR data from Feliz Natal – MT municipality was used. From the pointcloud, the digital surface model was generated with 0,5 meters spatial resolution. The trees were identified through the selection of the altitude outliers on the pointcloud. From the identified trees it were selected the border pixels and from the it was extracted altitude, general, plan and profile curvature values, as well as slope. The kmeans algorithm clustered the variables in 30 different combinations to find the best combinations to segmentate the identified trees. The group classified with nothing but altitude presented clear disposition of the points in two separate groups, whilst the other combinations of attibutes promoted confusion on the data classification. The number of pixels generated from the segmentation process was below the truth, indicating loss of information on the segmentation. However, the methodology was propitious to segment irregular shaped tree crown as the ones found in the transition Cerrado-Amazônia. Adjustment to the methodology are necessary for improving the accuracy of the segmentation process.