—Benjamin Kamphaus, ITT Visual Information Solutions
Despite advances in remote sensing technology, automated processes for classifying urban land use remain limited in scope and power. This study sought to address issues limiting present approaches by using buildings as an alternative unit of classification. Basic Formal Ontology (BFO) was employed to identify existing classification problems and suggest improvements. LiDAR data from the City of Austin were used for building extraction, and an impervious surface layer created from orthophotography and NDVI in ENVI was joined with the LiDAR data for the building classification. The building extraction method combined segmentation with a neural network classification to achieve a segment level accuracy of 92%, with a building detection rate of 74%. A land-use classification was applied to ground truth buildings with an accuracy of 80%, and to the extracted buildings with an accuracy of 79%.