Image pre-processing and classification
Satellite image pre-processing is an improvement of satellite data that
suppresses unwanted distortions and main objective to generating a more
direct linkage between the obtained data and real world phenomena
(Coppin et al., 2004; Butt, et al., 2015). Satellite data were
preprocessed in ERDAS imagine 14 software for layer stacking,
mosaicking, and subsetting of the image on the basis of Area of Interest
(AOI). All image data were assessed by allocating per-pixel signatures
and distinguishing the Islamabad capital into five classes with
reference to specific Digitial Number (DN) value. The identified land
cover classes were built-up area, agriculture, water bodies, forest and
barren land (Table 2). For the individual pre-determined land cover
type, training sites were identified by demarcating polygons around
representative sites. In this study, for supervise classification thirty
spectral signatures for each class were taken from the each satellite
imagery by using the pixels enclosed by these polygons. According to Gao
and Liu (2010) an adequate spectral signature is the one ensuring that
there is minimal confusion among the land covers to be mapped. The
maximum likelihood algorithm was applied for supervised classification
of the images. Supervise classification is mainly controlled by the
analyst as the analyst choose of the pixels which demonstrate the
desired classes. For medium spatial resolution data, such as Landsat
mixed pixels are a common problem, especially for the urban surfaces are
a mixture of features such as build-up land, grass, trees, roads and
water. For the improvement of classification accuracy and to produce
quality of land LULC maps, visual interpretation is very important.
Therefore, with help of field observations, reference data as well as
local knowledge can improve the results acquired using the supervised
algorithm (Jensen and Im, 2007; Butt et al., 2015).