In: Wagner W., Szekely, B. (eds.): ISPRS TC VII Symposium - 100 Years ISPRS, Vienna, Austria, July 5-7, 2010, IAPRS, Vol. XXXVIII, Part 7B
URBAN IMPERVIOUS SURFACE EXTRACTION FROM VERY HIGH RESOLUTION
IMAGERY BY ONE-CLASS SUPPORT VECTOR MACHINE
P. Li, H. Xu, S. Li
Institute of Remote Sensing and GIS, School of Earth and Space Sciences, Peking University, Beijing 100871, P R
China - pjli@pku.edu.cn
KEY WORDS: impervious surface, multi-level segmentation, One-Class SVM, very high resolution imagery, land cover
classification
ABSTRACT:
This paper proposes a new method for extracting impervious surface from VHR imagery. Since the impervious surface is the only
class of interest (i.e. target class), the One Class Support Vector Machine (OCSVM), a recently developed statistical learning
method, was used as the classifier. Rather than use samples from all classes for training in traditional multi-class classification, the
method only requires samples of the target class for training. The classification was conducted on object level. The proposed
method was evaluated and compared to existing methods using Quickbird image from Beijing urban area. The results showed that
the proposed method outperformed the existing method in term of classification accuracy. The method provides an effective way
to extract impervious surface from VHR images.
1. INTRODUCTION
Impervious surface is defined as any materials that water
cannot infiltrate, and has been recognized as an important
indicator in urban environmental assessment and valuable input
to planning and management activities (Lu and Weng, 2009;
Yuan and Bauer, 2006). The extraction of impervious surface
from remote sensing imagery has continued to be an important
problem for more than three decades. In recent years, the
increasing availability of very high resolution (VHR) imagery,
such as IKONOS, Quickbird and GeoEye-1, provides great
opportunity for detailed impervious surface mapping in urban
areas. Although some methods using VHR images have been
developed (Lu and Weng, 2009; Yuan and Bauer, 2006; Goetz
et al., 2003; Cablk and Minor, 2003; Zhou and Wang, 2008;
Roeck et al., 2009), obtaining highly accurate land cover and
impervious surface information from VHR imagery remains
challenging, thus new methods and techniques are still required.
However, since there is extensive occurrence of shadows in
VHR imagery caused by high buildings and trees in dense
urban areas, which leads to the reduced or total loss of spectral
information in the shaded areas, an important problem to be
addressed is to identify the impervious surfaces in shaded areas
(Lu and Weng, 2009). As in general land cover classification of
urban areas using VHR images, object based methods are also
commonly used to extract impervious surfaces (Cablk and
Minor, 2003; Yuan and Bauer, 2006; Zhou and Wang, 2008).
2. METHODS
In this study, we adopted a two-stage object based method to
extract impervious surface. At the first stage, shadow areas
were identified at object level generated by image segmentation.
At the second stage, shadow areas and non-shadow areas were
separately classified to extract impervious surface, using
one-class Support Vector Machine (One-class SVM or
OCSVM).
Prior to these two stages, multilevel hierarchical segmentation
using the proposed method was first carried out, different levels
of segmentation results were then selected for each stage. For
example, since shadow extraction at the first stage and shadow
classification at the second stage require different levels of
segmentation detail, shadow extraction was conducted at a
coarse level of segmentation, whereas the shadow classification