Goals and Background:
The goals of this lab are to use various image-processing
tasks to our study of remotely sensed images by accomplishing the following
tasks:
1. Use
various methods to create a study area derived from a larger satellite image
scene.
2. Use
spatial resolution techniques to optimize an image for improved image
interpretation.
3. Use
radiometric enhancement techniques to enhance image quality for improved image
interpretation.
4. Link
images in Erdas to Google Earth, thereby taking advantage of Google Earths
ability to be a high-resolution image interpretation key for ancillary
information.
5. Introduce
resampling techniques.
6. Explore
the differences in Erdas Mosaicking tools by examining the tool’s output.
7. Introduce
Binary change techniques, as well as using various GIS platforms to interpret
those changes.
Methods:
There are two ways to create an Image Subset in Erdas,
by using the Inquire box and by using a shapefile. In order to create an
inquire box you can right click on a satellite image in Erdas and select the
inquire box option. The inquire box can then be moved and manipulated to the
specific area of interest that you would like to subset. Once the area has been
determined, using the Subset and Chip tool and selecting Create Subset Image
will allow you to make any final adjustments and to finalize the creation of
the Image subset (Figure 2).
To create an Image Subset with a shapefile is a very
similar process. By overlaying a shape file over a satellite image, we can use
the same process of creating an Image Subset by using the Subset and Chip tool
and creating an output file of the Image Subset with the same boarders of the
shapefile (Figure 3).
Pan sharpening is a great way of improving image
quality for interpretation. By taking a image and merging it with a higher
resolution image results in increased clarity while still retaining the original
images data. In Erdas, using the raster tool, “Resolution merge” two images can
be pan sharpened using a multiplicative algorithm and the Nearest Neighbor resampling
technique, a higher resolution image is produced (Figure 4).
Another problem often encountered by analysts is atmospheric
haze, which results in cloudy images. In order to correct haze, a radiometric
enhancement technique is used, known as Haze Reduction. This technique brings
out the true color of the satellite image and results in a clearer image (Figure 5).
Linking Google Earth to a satellite image in Erdas can
serve to provide the analyst with extra information in order to undergo the image
interpretation process to fully intemperate images in remote sensing. Google
Earth can provide this information easily, as Google Earth uses up to date,
GeoEye high-resolution satellite data. Simply by clicking on the Connect to Google
Earth option, and by selecting sync views an analyst can have a very powerful
image interpretation key (Figure 6).
If an analyst is required to increase or decrees the
size of an image’s pixels, a resampling technique will be required. By
resampling up, the size of the pixels will decrease, while resampling down will
reduce the size of the pixels. By selecting “Resample Pixel Size” in the
spatial tools of the raster tab in Erdas, an analyst can determine both the
resampling technique required and how those pixels will be interpolated. Different
interpolation methods have positives and negatives (Figure 7).
An additional issue a remote sensing analyst may
encounter is that the study area that the analyst is interested in may be too
large to be represented by a single image of a satellite data. To fix this
issue the analyst must mosaic two images together, creating a larger scene to analyze.
Erdas presents two options for image mosaicking, Mosiac Express and Mosaic Pro.
Both options require the image files to be mosaicked to be added in a
particular way, before the images are added clicking the multiple tab, and then
selecting the radio button multiple image in virtual mosaic must be completed
prior to adding the images or the mosaic tools will not run. Once added the analyst
will need to select the Mosiac Express either tool or the Mosaic Pro tool to
run the mosaicking options. Mosaic express can be ran with little input,
however Mosaic Pro needs many options specified in order to complete a more accurate
image mosaic (Figure 8 and Figure 9).
Remote sensing can
be used to analyze the change in land cover over time by using a powerful model
known as Binary Change Detection or image differencing. By comparing the
brightness value of two remote sensing images taken years apart, an analyst can
model the differences in land cover that have occurred. By using the two input
operators interface in Erdas, an analyst can add images from many years apart
and run the model to see the change in land cover. An analyst can also estimate
which data will change by determining thresholds and looking at the images
histogram, to see which data values will change. In this case the rule of
threshold is used to establish the point at which the data will change. Using
the mean of the image values taken from the metadata, and then by adding 1.5(Standard
Deviation of the mean), the analyst can determine the upper threshold, or the
lower limit of the data change on the histogram. Similarly by subtracting 1.5(Standard
Deviation of the mean) and then making the number negative, the lower limit or
the upper boundary of the data change on the histogram will be determined (Figure
1).
Figure
1. Binary Difference Calculation.
Histogram depicts the original image, before processing. The red dashed lines
indicate the Lower Change Threshold and Upper Change Threshold calculated by
the mean ± 1.5 (standard deviation), with the Lower Change Threshold always
negative in value. Changes to the histogram when undergoing the binary
difference process will take place on the left and right of the thresholds
respectively.
The analyst can use model maker do
the calculation for them by simply subtracting the newer image from the older
image and adding the constant. The output of the model can then be put in
another model, which will create a binary image of the changed pixels by using
a conditional argument, EITHER 1 IF (the
output image > change/no change threshold value) or 0 OTHERWISE. By importing
that output into ArcMap an analyst can create a map showing the land cover
change over time (Figure 10)
Results:
The various methods to create a study area derived
from a larger satellite image scene can be seen below. Both subsets are of the boundaries
of the inquire box or the shape file respectively. In both cases the result is
a smaller image which contains just the area of interest.
Figure
2. Subsetting with the use of an Inquire box. Left
Viewer, Original image of the Eau Claire area from 2011 with Inquire box
(gold). Right Viewer, Eau Claire subset image extracted from the Eau Claire
area image, boundaries consist of the inquire box from the left viewer.
Figure 3. Subsetting with the use of a shapefile. Left Viewer, Original image of
the Eau Claire area from 2011 with an overlaid
shapefile. Right Viewer, New subset image whose boundaries consist of the AOI
from the Eau Claire shapefile.
The results of the Pan sharpen technique results an
image (Figure 4, right) which now has
the same pixel size as the panchromatic band (15mx15m) and has darker tones and
colors to aid in the image interpretation process.
Figure
4. Image Fusion. The left viewer
contains an image of Eau Claire and Chippewa counties in its original form. The
right viewer contains the same image that has been pan sharpened.
Similar to the Pan
Sharpening process, the Haze reduction process also results in a truer image
color to help in the image interpretation process, however haze reduction does
not resize the images pixels.
Figure
5. Radiometric Enhancement. Left
Viewer, original image (zoomed in) of the city of Eau Claire. Right viewer,
image after undergoing haze reduction.
By linking Erdas
to Google Earth we can see just how powerful of a selective key Google Earth can
be. The Image below displays a zoomed in image of down town Claire displayed
both on the
typical view of a satellite
band image and what that view looks like in pictures on Google Earth. As far as
image interpretation goes, Google Earth can be an extremely valuable tool.
Figure
6. Linking Image Viewer to Google
Earth. The viewer on the left displays down town Eau Claire. The viewer on the
right is linked to google Earth and displays downtown Eau Claire at a linked
distance. This showcases how useful Google Earth can be in the image
interpretation process.
From a large scale,
it is very hard to tell the difference in the original image to the interpolated
image. Both the Nearest Neighbor and Bilinear Interpolation image results are indistinguishable
from their original images at this scale. However when viewed from a small scale
the differences between the interpolation methods is striking. The Nearest
Neighbor Interpolation derives cell values from the closest cell of the
original image to the closest cell of the new image, which results in most
cells values being similar in the interpolated image as the old image. However,
the bilinear interpolation averages the cell values from the 4 closest cells of
the old image to the interpolated image. This results in a very “blury” interpolated
image where the cells begin to blend in to one another. (here
is an example of what both interpolation methods look close up)
Figure
7. Resampling Up. The left viewer
shows the original image (30mx30m) compared to the right viewer, which has the
Resampled Up image (15mx15m). The method of Interpolation is Nearest Neighbor,
which means that the values for the cells in the new image were determined by
the closest cells of the old image.
While Mosaic Express is faster and not as computationally
intensive as Mosaic Pro, the results are of both tools are drastically
different. Mosaic Express leaves the distinct boarder of both images intact and
is unable to adjust the images to have similar color outputs. Mosaic Pro is
able to do both of these functions, however Mosaic Pro has more inputs required
from the analyst in order to complete the operation. Despite the extensive
inputs Mosaic Pro is the preferred choice for mosaicking in image
interpretation because of the higher quality of the output image produced.
Figure
8 and Figure 9. Comparison of Image Mosaicking
Tools from Erdas: Mosaic Express and MosaicPro. The viewer on the right is
displaying two images mosaicked with the Mosaic Express tool. The viewer on the
right is displaying two images mosaicked with the Mosaic Express tool.
After completing the calculations
of the binary change detection, the image can be loaded into ArcMap and turned
into a proper map, with base data to put the land cover changes into context.
If we look at the distribution of land cover change, the largest changes take
place in areas around urban areas, such as small towns and suburbs. Further analysis
may reveal that this is due to population shifts where people are moving towards
more urban areas. Alternatively, it may be that there has been an increase in
forested areas being converted to farmland throughout the area during this time.
Although the actual answer may be neither
hypothesis, the model can depict many stories caused by the changes in land
cover over time.
Figure
10. Mapping Pixel changes via spatial modeler and ArcGIS. Map depicts change
in pixel values from August 1991 and August 2011. Green areas Indicate pixels
which changed.
Sources:
Extension: Remote Sensing
Resampling Methods, Retrieved 11/1/2016.
“What Are you Looking at?”
Discovery Station. Retrieved 11/1/2016.
http://www.smithsonianconference.org/climate/wp-contents/uploads/2009/09/ImageInterpretation.pdf
Satellite images are from Earth
Resources Observation and Science
Center, United
States Geological Survey. Shapefile is from Mastering ArcGIS 6th edition Dataset
by Maribeth Price, McGraw Hill. 2014.
Change in Land cover from
1991-2011 Map Sources are from Esri, Delome, Gebco, NOAA, NGDC
and other contributors. Cartography by Paul Cooper
No comments:
Post a Comment