The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences
Download
Publications Copernicus
Download
Citation
Articles | Volume XLI-B7
https://doi.org/10.5194/isprs-archives-XLI-B7-981-2016
https://doi.org/10.5194/isprs-archives-XLI-B7-981-2016
14 Oct 2016
 | 14 Oct 2016

THE USAGE OF RUSBOOST BOOSTING METHOD FOR CLASSIFICATION OF IMPERVIOUS SURFACES

M. H. Kesikoglu, U. H. Atasever, C. Ozkan, and E. Besdok

Keywords: RUSboost, Classification tree, Impervious surface, Image classification, Remote sensing

Abstract. Impervious surface areas are artificial structures covered by materials such as asphalt, stone, brick, rooftops and concrete. Buildings, parking lots, roads, driveways and sidewalks are shown as impervious surfaces. They increase depending on the population growth. The spatial development of impervious surface expansion is necessary for better understanding of the urbanization status and its effect on environment. There are different impervious surface determining approaches met in literature. In this paper, it is aimed to extract the impervious surface areas of Kayseri city, Turkey by using remote sensing techniques. It is possible to group these techniques under a few main topics as V-I-S (vegetation-impervious surface-soil) model, based on spectral mixture analysis or decision tree algorithms or impervious surface indices. According to these techniques, we proposed a new technique by using RUSBoost algorithm based on decision tree in this study. In this scope, Landsat 8 LDCM image belonging to July, 2013 was used. Determining of impervious surface areas accurately depends on accuracy of image classification methods. Therefore, satellite image was classified separately by using Classification Tree and RUSBoost boosting method which increases accuracy of the classification method based on decision tree. Classification accuracies of these supervised classification methods were compared and it was observed that the best overall accuracy was obtained with RUSBoost method. For this reason, RUSBoost method was preferred to determine impervious surface areas. The overall accuracies were obtained 95 % with Classification Tree and 97 % with RUSBoost boosting method.