Building type classification using spatial and landscape attributes derived from LiDAR remote sensing data

• Published in: Landscape and urban planning 2014-10, Vol.130, p.134-148
• Main Authors: Lu, Zhenyu, Im, Jungho, Rhee, Jinyoung, Hodgson, Michael
• Format: Article
• Language: English
• Subjects: Animal, plant and microbial ecology; Applied ecology; Biological and medical sciences; Building classification; Buildings; Classification; Conservation, protection and management of environment and wildlife; Construction; Decision trees; Forestry; Fundamental and applied biological sciences. Psychology; General aspects; General aspects. Techniques; General forest ecology; Generalities. Production, biomass. Quality of wood and forest products. General forest ecology; Houses; Landscapes; LiDAR; Machine learning; Random forest; Remote sensing; Support vector machines; Teledetection and vegetation maps; Urban planning
• ISSN: 0169-2046; 1872-6062
• DOI: 10.1016/j.landurbplan.2014.07.005

•We perform building type classification using spatial and landscape attributes derived from LiDAR remote sensing data.•We compare four machine learning methods for building type classification.•We examine the importance of the spatial and landscape variables for building usage identification. Building information is one of the key elements for a range of urban planning and management practices. In this study, an investigation was performed to classify buildings delineated from light detection and ranging (LiDAR) remote sensing data into three types: single-family houses, multiple-family houses, and non-residential buildings. Four kinds of spatial attributes describing the shape, location, and surrounding environment of buildings were calculated and subsequently employed in the classification. Experiments were performed in suburban and downtown sites in Denver, CO, USA, considering different building components and neighborhood environments. Building type classification results yielded overall accuracy>70% and Kappa>0.5 for both sites, demonstrating the feasibility of obtaining building type information from LiDAR data. The shape attributes, such as width, footprint area, and perimeter, were most useful for identifying building types. Environmental landscape attributes surrounding buildings, such as the number of road and parking lot pixels, also contributed to obtaining building type information. Combining shape and environmental landscape attributes was necessary to obtain accurate and consistent classification results.