Research Methodology - Historical Land Use/Land Cover Classification Using Remote Sensing - page 61

Chapter 5
Research Methodology

This chapter gives a review about techniques and methodologies that were applied
to answer the presented research questions and to confirm the hypothesis of this
thesis. The conceptual workflow chart of the thesis is illustrated with an overview
provided in Fig. 5.1 .
Tone or color is the basis factor for most methods of digital image analysis. It is
represented as a digital number in each cell of the recorded remote sensing image.
The first step is applying a various procedures of preprocessing on the raw digital
image. To carry out image classification, many steps have to be considered:
choosing of a fit classification system; choosing of training samples; preprocessing
of image(s); drawing out the feature; choosing of fit classification approaches;
processing the resulted products of classification; and accuracy assessment.
Utilization of several variables during the classification process can make the
classification accuracy worse because of unlike capabilities in separation between
classes of interest (Price 2003 ). Therefore, many potential variables were used in
image classification for the study case of this thesis, including spectral signatures,
vegetation indices and transformed images (NDVI), multi-temporal images (1975,
1987, 2005 and 2007; April, May, July and August), multi-sensor images and
ancillary data (GPS measurements, spectral information, statistical records,

Google Earth etc.).
In this thesis, I will try to propose the methodological means which contribute
to analysis of various data and information, and to integrate some of these data
between each other, if necessary, to extract the information/results from the
satellite images, to be presented in the final thematic maps.
Setting three local levels with multi-temporal levels to process sensory data
available for obtaining thematic maps.
The first local-level: this level was embodied in the four administrative areas'
borders (Menbij, Al-Jurnia, Ain Eisa and Athawra), and was accredited to test and
compare several algorithms and automated classification methods in order to best
determine the optimized algorithm and method of classification. Algorithms such
as MLC, NN and SVM were tested in two ways. The first approach relied on a
hierarchical shape and involved the extraction of classification outcomes through