Detection of roof materials based on an object-oriented approach using worldview-2 satellite imagery

One of the most challenging tasks in urban remote sensing is detection of impervious surface (IS) which plays an important role in assessing urban environmental conditions. However, accurate impervious surface extraction is still a challenge. New methods are needed due to the rapid expansion and development of urban centers which are able to do more frequent updating of existing databases instead of traditional methods. In this study, detection of the IS especially the roof surfaces based on their materials is proposed. Detection of roof types and conditions are important and the information on the roof material types can be useful for different applications such as disaster preparedness, urban heat island assessment and runoff quality. Due to the limitations of airborne hyperspectral data in which data acquisition is normally expensive, the coverage area is limited, and the analysis can be too complex, very-high-resolution (VHR) imagery, such as Worldview-2 (WV-2) image was used. In order to do supervised classification and extract the IS at the parcel level, especially roof materials, adequate training data are needed, but lack of sufficient training data is one of the limitations, due to the security of buildings, permission to access the roof could not be obtained or access was impossible. The Object-oriented (OO) approach was used in order to utilize the spectral, spatial and textural information which are inherent in VHR imagery. In order to define the objects based on OO approach, certain rules should be defined. This is a difficult task due to the requirement of the prior knowledge about the objects. Lack of generic rule to extract the IS is another limitation in urban remote sensing. The main goal of this research is to build the generic rules based on the spectral, spatial and textural information to predict roof materials in WV-2 images without using training data. A generic model is proposed that is based on spectral, spatial and textural information which were extracted from available training data. Furthermore, discriminant analysis (DA) is used for dimensionality reduction and to discriminate between different spatial, spectral and textural attributes. The generic model consists of a discriminant function based on linear combinations of the predictor variables that provide the best discrimination between the groups. The DA result shows that of the 54 attributes extracted, only 13 attributes related to spatial, spectral and textural information are useful for discriminating different roof materials. Finally, this model was applied to different WV-2 images from different areas and proved that this model has a good potential to predict roof materials in different study areas with more than 81% accuracy. This is performed on WV-2 images without using training data