Erdas Imagine possess many effective functions built in for the analysis and altering of satellite and image data. The problem is comes from learning to use this built in software. Th purpose of lab four was to introduce students to a number of these built in functions and allow for the familiarization of the use of these functions. The methods and functions that were required for students to learn and use were as follows: delineating a study area from a larger source image, demonstrating how spatial resolution of images can be optimized for visual interpretation, introducing radiometric enhancement techniques, linking a satellite to Google Earth to use as a sourcing image, introducing a variety of resampling methods, examining and utilizing image mosaicking, and exploring the use of binary change detection with graphical modeling.
Methods
Part 1: Image Subsetting
In part one, image subsetting was the primary focus, both by using an inquire box and by utilizing an area of interest. First, the base image "eau_claire_2011.img" was brought into an Erdas image view. With this image maximized in the view, an inquire box was generated around the Eau Claire - Chippewa County area of the image. Then, the Subset & Chip tool "Create Subset Image" in the raster menu was launched. This would generated an output image from the original image, "eau_claire_2011.img", marked as the input file. The output image was selected to be created "From Inquire Box". This meant that the output image would be created only what was in the inquire box located on the original image, the Eau Claire - Chippewa County area. The "Create Subset Image" box was then run with the remaining default parameters, and the output save as "eau_claire_2011sb_ib.img". This was the subset image created from the inquire box. Next, the view was cleared of all bu the original "eau_claire_2011.img". Then the shapefile "ec_cpw_cts.shp" was added to the viewer. Both areas of the shapefile were selected with the shift key. The "paste from selected object" was then utilized in the "Home" menu, creating an area of interest around the shapefile. This area of interest was then saved as an "AOI Layer" under the "Save As" menu and titled to create the "ec_cpw_cts.aoi" area of interest layer. Next, the "Create Subset Image" from before was launched, once again utilizing the original image as the input file. The difference was the "AOI" option was used in order to utilize the created AIO layer from before. The tool was then run and the output save as "ec_cpw_2011sb_ai.img", an image taken from the original of only the area of interest, the Eau Claire and Chippewa Counties.
Part 2: Image Fusion
This involved the creation of a higher resolution image from a lower resolution image in order to aid in visual interpretation. The 30 m spatial resolution "ec_cpw_2000.img" was brought into a view, with the higher 15 m resolution panchromatic "ec_cpw_2000pan.img" being brought into another view. Under the Raster tool menu, Pan Sharpen was selected, with the "Resolution Merge" tool being selected from the drop-down menu. In the "Resolution Merge" window, the panchromatic image was selected to be the High Resolution input image, and the original "ec_cpw_2000.img" was selected to be the multispectral input image. The method was set to "Multiplicative" and the Resampling Technique was set to "Nearest Neighbor". The resolution merge model was then run.
Part 3: Simple Radiometric Enhancement
This demonstrated the processes behind a radiometric technique designed to reduce haze on a target image. The "eau_claire_2007.img"image was brought into a Erdas viewer. This image showed significant haze in the bottom-right corner of the image. To reduce this, the raster processing tool "Haze Reduction", found under Radiometric tools was launched. The original image was marked as the input file. All default parameters were accepted, and the tool was run, with the output being saved as "ec_2007_haze_r.img". This was the haze reduced image, which was then compared to the original.
Part 4: Linking Image Viewer to Google Earth
This next segmented demonstrated how to link the image viewer directly with Google Earth, for quick comparison or to be used as an image interpretation key. The "eau_claire_2011.img" was brought into a viewer. From here, the Google Earth key was selected from the top of the Erdas interface, then the "Connect to Google Earth was selected. This brought up the Google Earth View. "Match GE to View" was then used to make the Google Earth viewer display the same area as the image in Erdas. Afterwards, the "Sync GE to View" tab was used to allow synchronized viewing of both Erdas and Google Earth.
Part 5: Resampling
In order to increase or decrease pixel size, resampling is required. This part of the lab detailed the processes behind resampling up (decreasing) the pixel size. The "eau_claire_2011.img" was moving into a viewer, and the pixel size was recorded from the metadata to be of 30 m. From here, the raster tool "Resample Pixel Size" was selected from the Spatial tools drop-down menu. "eau_claire_2011.img" was selected as the input image. The output cell size was changed from the original 30x30 meters to 15x15 meters and the resample method was set to nearest neighbor. All other parameters were left as default and the tool was run, with the output being saved as "eau_claire_nn.img". However, when compared to the original image, no visual change could be spotted. This was because the resampling technique nearest neighbor was dividing each of the original pixels into four new pixels without changing their color value, as it was applying the color of the original image where the new pixel were. to correct this, the "Resample Pixel Size" tool was run again, with the same parameters as before, just with the resampling method being changed to Bilinear Interpolation. This means that the new pixels being generated took their color based on their location from the original surrounding pixels, with the closest pixels lending more weight to the final color. The tool wars then run and the resample saved as "eau_claire_bli.img". This was compared to the original image, with the new image showing visual differences from the original.
Part 6: Image Mosaicking
Image mosaicking is a tool used to link or overlay multiple satellite images for visual interpretation when one satellite image isn't large enough to produce the desired results. First, the images "eau_claire_1995p26r29.img" and "eau_claire_1995p25r29.img" were individually added to a viewer. However, each image had special parameters under the "Multiple" and "Raster Options" selected before each image was added from the "Select Layers to Add" window. Under "Raster Options", Background Transparent and Fit to Frame were checked, and under "Multiple", the radio button "Multiple Images in Virtual Mosaic" was selected. This resulted in the images being displayed one overlayed on the other. Next, the raster Mosaic tool "Mosaic Express" was opened. From here, the "eau_claire_1995p25r29.img" was added first, followed by the "eau_claire_1995p26r29.img" in order to layer them correctly. All other options of the Mosaic Express tool were left as default, and the output was saved as "eau_claire1995msx.img" before running the model. After displaying the newly created image, it was discovered that the colors between the two images did not evenly blend on the overlap area. To make up for this, the "MosaicPro" raster Mosaic tool was launched. Each of the two original images was added to the tool, with the Image Area Options being changed to Compute Active Area for each image. Using the Select tool, the order of the images was experimented with until " eau_claire_1995p25r29.img" was the bottom image. This meant the other image would be the one overlayed over the top. The radiometric properties of the images were synchronized by checking the "Use Histogram Matching" tab under the Color Corrections tool. The "Set Overlap Function" was checked to see if the Overlay function was set as default, and the Mosaic process was run, with the output being compared to the original Mosaic Express output. The new MosaicPro image showed far better blending of color in the overlap area.
Part 7: Binary Change Detection
Using images taken of the same area at different times, it is possible to use pixel brightness to detect change in the images from one to the next. In one viewer, "en_envs1991.img" was displayed. In a second viewer, "ec_envs2011.img" was displayed. Next, the Raster tool "Function" was selected. "en_envs1991.img" was selected as input 2 and "ec_envs2011.img" was selected as input 1. The operrator was changed to "-". The Layer was changed from "All" to "4" to imply the change detection. Afterwards, the image differentiating process was run and the results displayed. Using the metadata, the histogram was displayed for the output. The cutoff point was determined to be the mean + 1.5(Standard deviation). This is entirely subjective but was determined to be the best cutoff point between areas of little change and areas of high change. Using this data, a histogram was made showing the center value of the histogram and the cutoff points were shown to be the center value of the histogram + or - (mean + 1.5(Standard deviation)). From here a model was constructed using the Model Maker tool, with the two input rasters being "ec_envs_2011_b4.img" and "ec_envs_1991b4.img" NIR images from the designated area, taken in 2011 and 1991, respectively. These input rasters were input into the function
"$n1_ec_envs_2011_b4 - $n2_ec_envs_1991_b4 + 127"
which would map the pixels which had changed between the two images. "127" was the determined constant of this function. The model output the function as a single raster, which was saved as "ec_91-11chg_b.img". This image was brought into the viewer, and its metadata was examined for a cutoff point. Is was decided from this that the threshold would be determined from "mean + (3 x standard deviation)". The Model Maker tool was launched again, but instead with a single input raster for a function which output as a single raster. "ec_91-11chg_b.img" was inserted into the input raster, and the function was determined to be a Conditional function of
EITHER 1 IF ( $n1_ec_91> change/no change threshold value) OR 0 OTHERWISE.
where threshold value = mean + (3 x standard deviation)
This meant that only areas of sufficient change would be displayed in the output raster. The output raster was saved as "ec_91-11bvis.img". Using ArcMap, this was displayed over the original NIR image "ec_envs1991b4.img" in order to create a aesthetically pleasing map that concisely displayed the changed areas of the displayed region from 1991 to 2011.
Results
Part 1
The results of part one demonstrated how an image can be broken down into smaller image subsets (Figure 1). Using image subsets can along for more effective image manipulation and interpretation, cutting out data not useful to the study. In this case, area of interest and inquire box tools were used in order to create image subsets of the Eau Claire and Chippewa County areas from a larger image.
Part 2
With the use of pansharpening, an output image could be created of the original input image with the greater resolution the panchromatic image. The pan-sharpened image now has the 15x15 meter pixel size of the panchromatic image (Figure 2). For visual interpretation, this presents an enormous advantage. For most satellite imagery, the panchromatic band can be recorded at a higher resolution than other spectral bands. By using pan-sharpening, the higher resolution can be applied to the other spectral bands without increasing the power of the satellite.
Part 3
By using haze reduction, the haze present at the bottom right corner of the screen has largely been removed (Figure 3). This technique proves to be very effective at cleaning up an image's haze. With this tool, images that were largely marred and unusable for image analysis can now be effectively used. It is interesting to note that while the haze has been cleared from the haze reduction image, the shadow of the haze appears to remain. This suggests this technique may not be perfect, and that it is still best to use a satellite image without haze unless absolutely necessary.
Part 4
By linking an image in an Erdas to Google Earth, attempts at visual interpretation have been greatly improved. Google Earth, when linked to a relevant satellite image, acts a high resolution Selective interpretation key. Google Earth displays the same geographical area as the linked Erdas view in the visible spectrum while also providing annotations and labels for many of the key landmarks and features displayed (Figure 4).
Part 5
Through the use of several resampling techniques, it is possible to increase the resolution of an image without the use of pansharpening. This is particularly helpful in a case where the original image may not have a accompanying panchromatic image with a higher resolution. However, it is important to note how an image should be correctly resampled. If the resampling method is set to nearest neighbor when resampling to a higher resolution, such as in this case where the image was resampled from a 30x30 meter resolution to a 15x15 meter resolution, the original pixels will merely be broken into smaller pixels of the same color as the original (Figure 5). This is because the nearest neighbor will of the new pixel will have always been the original pixel.
To correct this while resampling up to a higher resolution, it is important to change the resampling technique to Bilinear Interpolation. This will generate a resampled image with a higher resolution and a better blend of colors in the newly generated pixels (Figures 6 and 7). This is because the colors of the new pixels are generated based on their relative locations to all of the original surrounding pixels, with the closer original pixels lending more weight to the color of the new pixels.
Part 6
By use of image mosaicking techniques, multiple images can be combined using their overlap area to generate a linked image. This is particularly useful when the study area is larger than one satellite image. Depending on the mosaicking technique used, a simple overlay is generated, like that with Mosaic Express (Figure 8), or a more even transition of colors between the overlap of the images can be created using MosaicPro and Histogram Matching (Figure 9).
Part 7
With the use of binary change detection, changes in pixel value can be detected from one image to the next. By the use of the metadata, cutoff points can be determined when it comes to how much change is necessary in order for it to be displayed. A histogram can be found within the metadata and used to display the cutoff points (Figure 10), as is shown for "ec_91-11chg_b.img". Furthermore, using binary change detection and model maker, an image can be generated which shows these significant changes visually. This image file can then be constructed into a map to easily display this information. Figure 10 shows a map of generated from the finished product of a binary change detection process between the images "ec_envs_2011_b4.img" and "ec_envs_1991b4.img", being taken in 2011 and 1991 respectively. This map shows significant change around population centers, leading to the explanation that the change in images is due to urban expansion over the last 20 years, resulting in areas surrounding cities to be converted for residential use. Binary change detection is not limited to urban expansion analysis. It can also be used to detect deforestation and monitor land use.
Sources:Wilson, C. (2016). Lab 4: Miscellaneous Image Functions. Eau Claire, Wisconsin.
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
No comments:
Post a Comment