Comparison of supervised classification technique of landuse map using high resolution image / Muhammad Firdaus Mohammad Harun

The land cover relate with physical feature of land surface. Land cover can be
categories such as development area, vegetation areas, rural area, urban area and
anything rely on the land surface. Remote sensing have been used to detect the
changes of the land covers occurs by human activity. In this project, the objective is to
generate supervised classification SPOT 7, to determine the accuracy of classification
using maximum likelihood, minimum distance, mahalanobis distance and spectral
angle algorithm and to produce the land use map. The algorithm were used to perform
the supervised classification. The landuse were classified into six classes i.e. shrub,
forest, paddy, cropland, build up and water. The accuracy assessment using error
matrix method were done. A total of sixty (60) ground data were used to validate the
accuracy of the classification. The result shows that maximum likelihood algorithm
has the highest value for overall accuracy and overall kappa statistic which is 87%
and 84% respectively. The lowest value shows by minimum distance algorithm is
68% and 61% respectively.