Geography Reference
In-Depth Information
ratio; (3) remotely sensed data are captured in digital forms and can thus be easily
jointed with other types of ground feature information through such techniques as
GIS; and (4) the large economies of scale offered by digital satellite image pro-
cessing make it fairly low-cost to map large areas, meaning it is easier and more
cost effective to produce large amounts of map products.
Although the optical remote sensing systems such as LANDSAT-MSS/TM/
ETM+, ASTER, and SPOT have limitations in obtaining cloud-free imagery and
the resulted difficulties in performing spectral classification for specific categories
of LULC (Ulaby et al.
1982
), they have proven an efficient device for LULC-
mapping (Ji
2000
). Kanellopoulos et al.
1992
conducted a 20 class classification
test on SPOT High-Resolution Visible (HRV) images, and the end-result was
proven to be satisfactory. De Colstoun et al. (
2003
) applied a decision tree on
multi-temporal images from the ETM+ to distinguish between 11 features of land
cover. The overall accuracy was clearly enhanced by using classifier ensemble
techniques, as boosting. The paper from Berberoglu et al. (
2007
) aimed to evaluate
the usefulness of integrating texture measures into MLC and ANN classifications
in a Mediterranean environment, using LANDSAT-TM-imagery. The best clas-
sification accuracies were reached by using the ANN classifier. The dealing with
the measures of texture characteristics were most effectively with the ANN rather
than the MLC classifier. Yuan et al. (
2009
) explained and applying an automated
two-module ANN classification system, i.e. an unsupervised SOM network
module and a supervised MLP neural network module, using LANDSAT-TM.
After an evaluation of the performance of MLC, DA, and ANN in image classi-
fication, ANN classifications have the advantages in image accuracy overall and
for single land cover classes.
LULC-Classification using the three VNIR- and six SWIR- bands of ASTER-
data has been discussed in the past 10 years. The most commonly used approach is
separating the ASTER into two sets of images, i.e. 15 and 30 m resolution, where
each have three and six spectral bands, respectively. For each set, support vector
machine (SVM)-based algorithms (Zhu and Blumberg
2002
) or segmentation
algorithms (Marcal et al.
2005
) were applied for processing of classification. An
approach based on wavelet fusion was proposed by Bagan et al. (
2004
). Other
studies based on the Principal Component Analysis (PCA) were used to the nine
VNIR and SWIR. From the earlier obtained principal components, a supervised
MLC was implemented (Gomez et al.
2005
). But, most of the approaches referred
to have not adopted thermal band data (TIR) in classification processing. Jianwen
and Bagan (
2005
) used ASTER and the Kohonen's Self-Organized neural network
feature Map (KSOM) to LULC-classification. It classified 7 % more accurately
than MLC. Also, the study showed that the quality of ASTER was good for LULC
classification. Yüksel et al. (
2008
) used ASTER and converted it into Top Of
Atmosphere reflectance data (TOA) to generate LULC-maps according to the
CORINE-Land cover project, using supervised and the knowledge-based expert
classification systems to get a superior accuracy of the classified image.
These optical remotely sensed data can be integrated with recordings from
remote sensing active systems such as the microwave sensors (e.g., Synthetic
Search WWH ::
Custom Search