Support vector machine and object-oriented classification for urban impervious surface extraction from satellite imagery

One of the most important applications of remote sensing in urban area is impervious surface information extraction. Previous research has shown that satellite imagery has the potential and advantage for impervious surface estimating. In particular, the high resolution imagery, which has a spatial resolution in the meter to sub-meter range, is very useful for high accuracy mapping and monitoring of urban impervious surface. In order to extract the high resolution urban impervious surface accurately and effectively, an object-oriented classification method based on SVM is employed in this paper. The prominent advantage of object-oriented classification is that different shape and texture characteristics of objects can easily be calculated on the segments. Support vector machine (SVM) is a supervised machine learning method that performs classification based on the non-parametric statistical learning theory. In this study, a case study was conducted for impervious surface mapping in Beijing with WorldView-2 imagery. According to the experiment results, the combination of SVM and object-oriented has shown promise in improving the quality of impervious surface extraction, and the overall accuracy of 93.4% and kappa coefficient of 0.921 were achieved. In addition, owing to the fact of strong spectral confusion between some landcover types, which still makes high extraction errors of certain land covers. In order to improve the accuracy of impervious surface extraction, the integration of multi-source (LiDAR, hyperspectral remote sensing data) remote sensing data and multi-classifier will be the future direction.