Semantic classification of urban buildings combining VHR image and GIS data: An improved random forest approach

摘要: Abstract While most existing studies have focused on extracting geometric information buildings, only a few concentrated semantic information. The lack of cannot satisfy many demands resolving environmental and social issues. This study presents an approach to semantically classify buildings into much finer categories than those by learning random forest (RF) classifier from large number imbalanced samples with high-dimensional features. First, two-level segmentation mechanism combining GIS VHR image produces single objects at scale intra-object components small scale. Second, semi-supervised method chooses unbiased considering the spatial proximity intra-cluster similarity buildings. Third, two important improvements in RF are made: voting-distribution ranked rule for reducing influences classification accuracy feature importance measurement evaluating each feature’s contribution recognition category. Fourth, urban is practically conducted Beijing city, results demonstrate that proposed effective accurate. seven used work more helpful studying problems.