Goal
and Background
The goal of this lab
was to use different methods to view correlations and associations of bands of
Landsat. Correlation and association analyses help to eliminate data
redundancies. There were two main tools that were used for this lab. The first
was feature space plot which showed association between Landsat bands. Feature
space plots are images that show ellipses that are either have high association
which would show a narrow ellipse or low association, which would have a wide
ellipse. The next tool used was a model maker. The model maker is a tool for
correlation and creates a correlation table.
Methods
The first part of the
lab was to look at associations of bands using feature space plots. As stated
earlier feature space plots are ellipse images that measure association.
Feature space plots are raster based image analyses. If there is a narrow
ellipse then there is a large similarity between bands. If there is a wide
ellipse then there is a low association between bands. To avoid data redundancy
a low association is needed.
To view a feature space
plot ERDAS Image was used. In ERDAS, the route to feature space plots is
Raster>Supervised>Feature Space Image. An input image is used and in this
case it was the eau_claire_2007 image. The output was saved in the student
personal folder. After these simple steps 15 feature space plot images were
shown. The images were used to answer questions and view the associations
between bands.
The second part of the
lab was to create correlation tables using a model maker. Model Maker is a tool
found under the Toolbox tab in ERDAS. For this lab there were 3 main objects
that were needed. These were a raster object, a function, and a matrix object.
The raster object was simply a raster image. For the model there were 3 images.
These were an image of Eau Claire, Key West, Florida, and the Bengal Province
in Bangladesh. The raster image is needed for the bands within the image. To
input a raster image it needed to be dragged and dropped into the model window.
These images were found in the Q-drive. 3 models needed to be run separately for
3 separate raster images.
The second object was the function. The
function used here was a correlation. To use the function it is a similar
process to input the raster image. Once in the model the function was selected
and the specific function “CORRELATION [ <raster>, IGNORE <value>]
was chosen. Replace the “<raster>” with the raster image “$n1_eau_claire_2007.
Also, <value> must be substituted with the number 0. This allows for the
correlation of this image and all bands.
The final object was the matrix object. The
matrix object is the output of the Model Maker. The matrix object was saved in
the Q-drive and was saved as .mtx file. For the Model Maker to work the 3
objects must be connected so that they know to interact with one another.
The 3 image correlation
tables were opened with Notepad and then transferred into Excel to allow for
correlation tables.
Results
 |
| A narrow ellipse showing a high association between bands |
 |
| A wide ellipse showing a low association |
