International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B2. Istanbul 2004
can be modified, therefore they depend on a series of factors,
such as: state of the sea, oil presence in the water, position of
the ship in relation to the aimed of the satellite, incidence angle
of the image, amongst others. Besides these characteristics, in
the case of ships in movement, the probable targets can also be
identified for the existence of its wake, that is to say, the
superficial waves caused by its movement.
In this work we will show the result of the use of the techniques
above metioned maids with the purpose of ships detection. For
this we will use the software described by Rocha et al (2001)
and Rocha and Stech (2003). In this software routines based on
the methodologies above cited had been implemented, being
accomplished studies regarding the values of the statistical
parameters of each distribution. After the observation of the
results obtained in diverse tests using diverse values for such
parameters, it was decided to use fixed values for each
statistical parameter in accordance with each distribution. Thus
each distribution had its statistical parameters established in
order to provide the best adaptation of the curve of its
histogram. In the case of the methodology described by
Eldhuset (1996) it was made only a change in the value used as
factor of comparison by this author.
After the act of receiving of the image, your digital processing
was initiated using Geomatica software. The image was,
initially, analyzed visually. After that visual analysis the image
was converted of 16 bits for 8 bits. Soon after the image was
registered using the ephemrides data that follow the image.
Made this, the image was exported to the “raw” format so that it
could be processed in software in question. Then, we made a
subset of the image of 2500 for 2500 pixels with the purpose to
reduce the computational effort.
In this software, the image was processed in order, in a first
stage, to discover which the value of grey level corresponds to a
probability of 99,5% of pixel to be a target. Discovered this
value of threshold, the value of the grey level of each pixel of
the image is compared with it and, in case that it is superior, it
will be associated in the resultant image to one pixel with value
255, that is to say, white. Otherwise it will be associated with
one pixel with value 0, generating black pixels in the exit
image. This way, the generated image will show pixels with a
minimum probability of 99,595 to be a target with the white
color and pixels with probability lower than 99,5% to be a
target with the black color. Only in the case of the use of the
methodology described by Eldhuset (1996) this procedure is not
obeyed, therefore in this case we worked with a 2x2 mask
covering the original image and generating an intermediary
image, that is compared with a factor of comparison described
by the author.
3. RESULTS AND DISCUSSIONS
Remote Sensing is a tool wide used at the present time for
diverse activities. In the case of ship detection, amongst other
areas, the use of SAR images has been highly used with great
efficiency. The fact of it could be acquired independent of the
clouds covering or the conditions of illumination, that is, as
much of day as the night, is a great big shot of these images.
The existence of speckle can make it difficult, but it does not
make impracticable the use of these images for this application.
The noise speckle is inherent to the process of formation of
SAR images, not being able therefore to be discarded.
However, the use of statistical distributions comes being used
with the purpose to allowing the digital processing of SAR
images in order to minimize the influence of speckle, without
with this causing the loss of information. In this work we can
126
verify the mentioned sttement. In Figure | we can see the
original image. This image was processed in the software
described by Rocha et al (2001) and Rocha and Stech (2003)
and, as we can see in Figures 2 to 7, the diverse tested
methodologies presented excellent results, only occurring small
inherent. variations to the characteristics of the used statistical
distributions.
n
Figure 1. Original image
The image generated by K distribution, shown in Figure 2, went
to that presented a bigger amount of noise in the oceanic area
and a bigger answer in the land area.
x Ud n =
Figure 2. K distribution processed image
These noises are represented in the image as diverse white
points in the oceanic area, could cause conflict in the
interpretation, because they could be interpreted as false targets.
Such fact happens by virtue of threshold established by
software for this distribution, as cited previously, to have been
established in a value not very high.
As we can see in Figures 3 and 4, the images generated by the
processing based on the use of the distributions Rayleigh and
Weibull, respectively, presented very good results, without the
presence of noises in the oceanic area, as it occurred with
distribution K. The image generated for the Weibull distribution
still presented a good reply in the land area.
Internation
Fig
Fig
Already in
by the pr
distributior
these distr
answer so 1
distributior
detecting tl
The Figure
on the use
This methc
detect the
area, presc
targets.
Fig