Distinguishing Façade Material Change using Hyperspectral Imaging

Publication Type:

Conference Abstract







To conduct city-scale computational modelling for infrastructure planning, micro-climate analysis, and disaster
mitigation, not only must the geometry of the built environment be detectable automatically but the component
materials must be as well. While extensive work has been undertaken for geometric recognition and feature
detection on buildings (e.g. VO et al., 2015), relatively little has been done for material identification (ZHU &
WOODCOCK, 2014). Furthermore, most of that work has been for the classification of urban land cover, with
extremely limited analysis applied to building material identification.
Today, remote sensing data in the form of hyperspectral imagery are widely used for identification of materials
in agriculture, environment, geology, astronomy, and more. Despite the widespread application of hyperspectral
imaging in many areas, this method has seldom been used in building material detection. As aerial hyperspectral
imagers do not get adequate information from building façades, close-range hyperspectral imaging, which is
a newer technique applied from the ground and recently used to study geological outcrops (KURZ et al.,
2013), can be applied to evaluate building façade material. This paper will investigate how different building materials can be differentiated using close-range remote sensing technology in the form of hyperspectral (near-infrared) data. For this purpose, the façade of a building containing multiple materials in Bergen, Norway, was scanned by a close-range, hyperspectral instrument. After masking non-building material, several pre-processing techniques were applied on the hyperspectral images including atmospheric and brightness correction, morphology effect removal, and bad pixel correction. Different materials on the building were then classified using supervised classification techniques (Linear Spectral Unmixing and Spectral Angle Mapper). The results showed the ability of hyperspectral data in the range of near-infrared to differentiate distinctive building materials.

Research area: