152
Space and Earth Remote Sensing: GIS Application to Environmental Impact Studies
density and to the presence of suspended particles), and the sensor operating
devices (rotating mirror, line or area CCD array).
It is absolutely necessary to perform an image georeference by determining the
geometric position in image and ground coordinates of a sufficient number of
ground control points; then, by using desirable geometric transformation
algorithms, to rectify the digital data through repositioning and radiometric
resampling.
2.3 Spectral signature extraction
If the remotely sensed data environmental analysis goal is to generate thematic
maps, the main task is to extract different spectral feature signatures in order to
classify the whole image.
A spectral signature is the radiant behavior (reflection or emission) of a certain
feature, this behavior being detected exactly where the feature is located. The
extraction has to be carried out by direct surveys, by means of spectrometers, or
by choosing some training samples on the multi spectral images.
A spectral signature is the minimum knowledge of the studied area that one can
obtain, and it is particularly important to understand how the analyzed
phenomenon are distributed on the ground and to know the relation with the
physical processes that generate them.
2.4 Definition of training samples
In order to compare the physical phenomena representation and the real evidence
of the same phenomena, it is necessary to extract some training samples that are
small portions of the studied area where a theoretical model can be built. Training
samples have to be both statistically meaningful and statistically representative,
because they are the only evidence one can distinguish.
2.5 Fully automated and supervised classification
Classification is the process that allows one to extract information from the raw
data.
Classification algorithms are essentially based on a radiometric distribution
statistical analysis of the electromagnetic signal in a whole data set (the digital
image). Such an analysis is performed both by evaluating the number of
radiometric classes in the digital image in order to select different data clusters
(unsupervised classification), and by comparing the likelihood of the radiant flux
that comes from the ground features collected from the sensor with the spectral
signatures extracted from the training samples.
The two different algorithm usages depend on a simple assumption; if the degree
of understanding of the studied area is low or non existent, it is necessary to
perform an unsupervised classification in order to extract the information (the
meaning of the different classes).
Space and Earth Remote Sensing: GIS Application to Environmental Impact Studies
density and to the presence of suspended particles), and the sensor operating
devices (rotating mirror, line or area CCD array).
It is absolutely necessary to perform an image georeference by determining the
geometric position in image and ground coordinates of a sufficient number of
ground control points; then, by using desirable geometric transformation
algorithms, to rectify the digital data through repositioning and radiometric
resampling.
2.3 Spectral signature extraction
If the remotely sensed data environmental analysis goal is to generate thematic
maps, the main task is to extract different spectral feature signatures in order to
classify the whole image.
A spectral signature is the radiant behavior (reflection or emission) of a certain
feature, this behavior being detected exactly where the feature is located. The
extraction has to be carried out by direct surveys, by means of spectrometers, or
by choosing some training samples on the multi spectral images.
A spectral signature is the minimum knowledge of the studied area that one can
obtain, and it is particularly important to understand how the analyzed
phenomenon are distributed on the ground and to know the relation with the
physical processes that generate them.
2.4 Definition of training samples
In order to compare the physical phenomena representation and the real evidence
of the same phenomena, it is necessary to extract some training samples that are
small portions of the studied area where a theoretical model can be built. Training
samples have to be both statistically meaningful and statistically representative,
because they are the only evidence one can distinguish.
2.5 Fully automated and supervised classification
Classification is the process that allows one to extract information from the raw
data.
Classification algorithms are essentially based on a radiometric distribution
statistical analysis of the electromagnetic signal in a whole data set (the digital
image). Such an analysis is performed both by evaluating the number of
radiometric classes in the digital image in order to select different data clusters
(unsupervised classification), and by comparing the likelihood of the radiant flux
that comes from the ground features collected from the sensor with the spectral
signatures extracted from the training samples.
The two different algorithm usages depend on a simple assumption; if the degree
of understanding of the studied area is low or non existent, it is necessary to
perform an unsupervised classification in order to extract the information (the
meaning of the different classes).