| Abstrak/Abstract |
Mapping and inventory of building objects are very important task in urban areas to provide baseline for census, urban planning, tax valuation and disaster mitigation. Therefore, a systematic, accurate, and repeatable method is required for identifying and mapping building objects. Building objects are relatively hard to map using pixel-based classification approach due to the high variation of building roof and shape. Object-based classification approach (GEOBIA) was developed to address the classification of complex object such as building, using spectral, spatial, morphology, contextual, and temporal aspect of the object. This study aims to (1) examine the effectiveness of spectral features in GeoEye-1 pan-sharpened image (0.5 m pixel size) to identify and map building objects, and (2) assess the accuracy of the mapping result. The location of the study sample was in parts of Padang City, West Sumatra, and the image used was GeoEye-1 acquired on January 2018. Image segmentation was done by multi-resolution segmentation method to delineate candidate segments for building objects. Each segment was then assigned into building and non-building classes by applying a rule-based classification algorithm. Several spectral features were incorporated in discriminating the objects, including several band ratio that involve all bands in GeoEye-1 (Blue, Green, Red and near-IR), iron oxide indices, mean value of red and NIR bands, border contrast of red and NIR bands, HIS, Quantile of the bands, etc. The map result indicates that building and non-building object could be separated using spectral features of GeoEye-1 image. However, there are some classification inaccuracy mainly for the densely populated urban areas where buildings objects are close to each other. An area-based accuracy assessment shows that the use of spectral features provides an overall accuracy of 68.7%. The results from this study show that (1) the selection of the right image segmentation parameters plays an important role in providing precise delineation of building objects from GeoEye-1 image, and (2) the use of spectral features only was not enough for classifying building and non-building in urban area. Future work will be aimed evaluate the role of geometry and contextual features, in addition to spectral features, in building extraction. |
